Topics in Mathematical Physics - Victor Palamodov
Topics in Mathematical Physics - Victor Palamodov
Prof. V.Palamodov
Spring semester 2002
Contents
Chapter 1. Dierential equations of Mathematical Physics
1.1 Dierential equations of elliptic type
1.2 Diusion equations
1.3 Wave equations
1.4 Systems
1.5 Nonlinear equations
1.6 Hamilton-Jacobi theory
1.7 Relativistic eld theory
1.8 Classication
1.9 Initial and boundary value problems
1.10 Inverse problems
Chapter 2. Elementary methods
2.1 Change of variables
2.2 Bilinear integrals
2.3 Conservation laws
2.4 Method of plane waves
2.5 Fourier transform
2.6 Theory of distributions
Chapter 3. Fundamental solutions
3.1 Basic denition and properties
3.2 Fundamental solutions for elliptic operators
3.3 More examples
3.4 Hyperbolic polynomials and source functions
3.5 Wave propagators
3.6 Inhomogeneous hyperbolic operators
3.7 Riesz groups
Chapter 4. The Cauchy problem
4.1 Denitions
4.2 Cauchy problem for distributions
4.3 Hyperbolic Cauchy problem
4.4 Solution of the Cauchy problem for wave equations
4.5 Domain of dependence
2
Chapter 5. Helmholtz equation and scattering
5.1 Time-harmonic waves
5.2 Source functions
5.3 Radiation conditions
5.4 Scattering on obstacle
5.5 Interference and diraction
Chapter 6. Geometry of waves
6.1 Wave fronts
6.2 Hamilton-Jacobi theory
6.3 Geometry of rays
6.4 An integrable case
6.5 Legendre transformation and geometric duality
6.6 Ferm at principle
6.7 The major Huygens principle
6.8 Geometrical optics
6.9 Caustics
6.10 Geometrical conservation law
Chapter 7. The method of Fourier integrals
7.1 Elements of symplectic geometry
7.2 Generating functions
7.3 Fourier integrals
7.4 Lagrange distributions
7.5 Hyperbolic Cauchy problem revisited
Chapter 8. Electromagnetic waves
8.1 Vector analysis
8.2 Maxwell equations
8.3 Harmonic analysis of solutions
8.4 Cauchy problem
8.5 Local conservation laws
3
Chapter 1
Dierential equations of
Mathematical Physics
1.1 Dierential equations of elliptic type
Let X be an Euclidean space of dimension n with a coordinate system
x
1
, ..., x
n
.
The Laplace equation is
u = 0,
.
=
2
x
2
1
+ ... +
2
x
2
n
is called the Laplace operator. A solution in a domain X
is called harmonic function in . It describes a stable membrane,
electrostatic or gravity eld.
The Helmholtz equation
_
+
2
_
u = 0
For n = 1 it is called the equation of harmonic oscillator. A solution is
a time-harmonic wave in homogeneous space.
Let be a function in ; the equation
, ) u = f,
.
=
_
x
1
, ...,
x
n
_
1
is the electrostatic equation with the conductivity . We have , ) u =
u + , u) .
Stationary Schr odinger equation
_
h
2
2m
+ V (x)
_
= E
E is the energy of a particle.
1.2 Diusion equations
The equation
u(x, t)
t
d
2
x
u(x, t) = f
in X R describes propagation of heat in X with the source density
f.
The equation
u
t
, p) u qu = f
describes diusion of small particles.
The Fick equation
t
c + div (wc) = Dc + f
for convective diusion accompanied by a chemical reaction; c is the
concentration, f is the production of a specie, w is the volume velocity,
D is the diusion coecient.
The Schr odinger equation
_
h
t
+
h
2
2m
V (x)
_
(x, t) = 0
where h = 1.054... 10
27
erg sec is the Plank constant. The wave
function describes motion of a particle of mass m in the exterior eld
with the potential V. The density [ (x, t)[
2
dx is the probability to nd
the particle in the point x at the time t.
2
1.3 Wave equations
1.3.1 The case dimX = 1
The equation
_
2
t
2
v
2
(x)
2
x
2
_
u(x, t) = 0
is called DAlembert equation or the wave equation for one spacial
variable x and velocity v.
The telegraph equations
V
x
+ L
I
t
+ R
I
x
= 0,
I
x
+ C
V
t
+ GV = 0
V, I are voltage and current in a conducting line, L, C, R, G are induc-
tivity, capacity, resistivity and leakage conductivity of the line.
The equation of oscillation of a slab
2
u
t
2
+
2
4
u
x
4
= 0
1.3.2 The case dim X = 2, 3
The wave equation in an isotropic medium (membrane equation):
_
2
t
2
v
2
(x)
_
u(x, t) = 0
The acoustic equation
2
u
t
2
, v
2
_
u = 0, = (
1
, ...,
n
)
Wave equation in an anisotropic medium:
_
2
t
2
a
ij
(x)
2
x
i
x
j
b
i
(x)
u
x
i
_
u(x, t) = f (x, t)
3
The transport equation
u
t
+
u
x
+ a (x) u b (x)
_
S(X)
(,
) , x) u(x,
, t) d
= q
It describes the density u = u(x, , t) of particles at a point (x, t) of
space-time moving in direction .
The Klein-Gordon-Fock equation
_
2
t
2
c
2
+ m
2
_
u(x, t) = 0
where c is the light speed. A relativistic scalar particle of the mass m.
1.4 Systems
The Maxwell system:
div (H) = 0, rot E =
1
c
t
(H) ,
div (E) = 4, rot H =
1
c
t
(E) +
4
c
I,
E and H are the electric and magnetic elds, is the electric charge
and I is the current; , are electric permittivity and magnetic per-
meability, respectively, v
2
= c
2
/. In a non-isotropic medium , are
symmetric positively dened matrices.
The elasticity system
t
u
i
=
x
j
v
ij
where U (x, t) = (u
1
, u
2
, u
3
) is the displacement evaluated in the tan-
gent bundle T (X) and v
ij
is the stress tensor:
v
ij
=
ij
x
k
u
k
+
_
x
j
u
i
+
x
i
u
j
_
, i, j = 1, 2, 3
is the density of the elastic medium in a domain X; , are the
Lame coecients (isotropic case).
4
1.5 Nonlinear equations
1.5.1 dimX = 1
The equation of shock waves
u
t
+ u
u
x
= 0
Burgers equation for shock waves with dispersion
u
t
+ u
u
x
b
2
u
x
2
= 0
The Korteweg-de-Vries (shallow water) equation
u
t
+ 6u
u
x
+
3
u
x
3
= 0
Boussinesq equation
2
u
t
2
2
u
x
2
6u
2
u
x
2
4
u
x
4
= 0
1.5.2 dimX = 2, 3
The nonlinear Schr odinger equation
h
u
t
+
h
2
2m
u [u[
2
u = 0
Nonlinear wave equation
_
2
t
2
v
2
_
u + f (u) = 0
where f is a nonlinear function, f.e. f (u) = u
3
or sin u.
The system of hydrodynamics (gas dynamic)
t
+ div (v) = f
v
t
+ v, grad v) +
1
grad p = F
(p, ) = 0
5
for the velocity vector v = (v
1
, v
2
, v
3
), the density function and the
pressure p of the liquid. They are called continuity, Euler and the state
equation, respectively.
The Navier-Stokes system
t
+ div (V ) = f
v
t
+ v, grad v) + v +
1
grad p = F
(p, ) = 0
where is the viscosity coecient.
The system of magnetic hydrodynamics
div B = 0,
B
t
rot (u B) = 0
u
t
+ u, ) u + grad p
1
rot B B = 0,
t
+ div (u) = 0
where u is the velocity, the density of the liquid, B = H is the
magnetic induction, is the magnetic permeability.
1.6 Hamilton-Jacobi theory
The Hamilton-Jacobi (Eikonal) equation
a
= v
2
(x)
Hamilton-Jacobi system
x
= H
(x, ) ,
= H
x
(x, )
where H is called the Hamiltonian function.
Euler-Lagrange equation
L
x
d
dt
L
x
= 0
where L = L(t, x,
.
x) , x = (x
1
, ..., x
n
) is the Lagrange function.
6
1.7 Relativistic eld theory
1.7.1 dimX = 3
The Schr odinger equation in a magnetic eld
h
u
t
+
h
2
2m
_
j
e
c
A
j
_
2
eV = 0
The Dirac equation
_
mI
_
= 0
where
0
= /t,
k
= /x
k
, k = 1, 2, 3 and
k
, k = 0, 1, 2, 3 are
4 4 matrices (Dirac matrices):
_
0
0
0
0
_
,
_
0
1
1
0
_
,
_
0
2
2
0
_
,
_
0
3
3
0
_
and
0
=
_
1 0
0 1
_
,
1
=
_
0 1
1 0
_
,
2
=
_
0
0
_
,
3
=
_
1 0
0 1
_
are Pauli matrices. The wave function describes a free relativistic
particle of mass m and spin 1/2, like electron, proton, neutron, neu-
trino. We have
_
mI
__
mI
_
=
_
+ m
2
_
I,
.
=
2
t
2
c
2
= 0,
= 4J
or F = dA, d dA = 4J
where J is the 4-vector, J
0
= is the charge density, J
= j is the
current, and A is a 4-potential.
7
Maxwell-Dirac system
= J
, (
+ eA
m) = 0
describes interaction of electromagnetic eld A and electron-positron
eld .
Yang-Mills equation for the Lie algebra g of a group G
F = , F
+ g [A
, A
] ;
F = J,
= J
gA
,
where A
is considered as a
connection in a vector bundle with the group G.
Einstein equation for a 4-metric tensor g
= g
(x) , x = (x
0
, x
1
, x
2
, x
3
) ; , =
0, ..., 3
R
1
2
g
R = Y
,
where R
,
+
.
=
1
2
(g
,
+ g
,
g
,
)
1.8 Classication of linear dierential opera-
tors
For an arbitrary linear dierential operator in a vector space X
a (x, D)
.
=
|j|m
a
j
(x) D
j
=
j
1
+...+j
n
m
a
j
1
,...,j
n
(x)
j
1
+...+j
n
x
j
1
1
...x
j
n
n
of order m the sum
a
m
(x, D) =
|j|=m
a
j
(x) D
j
, [j[ = j
1
+ ... + j
n
is called the principal part. If we make the formal substitution D ,
X
are called the symbol and principal symbol of the operator a. The
symbol of a linear dierential operator a on a manifold X is a function on
the cotangent bundle T
(X) .
If a is a matrix dierential operator, then the symbol is a matrix function
in X X
.
1.8.1 Operators of elliptic type
Denition. An operator a is called elliptic in a domain D X , if
(*) the principle symbol
m
(x, ) does not vanish for x D, X
0 .
For a s s-matrix operator a we take det
m
instead of
m
in this deni-
tion.
Examples. The operators listed in Sec.1 are elliptic. Also
the Cauchy-Riemann operator
a
_
g
h
_
=
_
g
x
h
y
g
y
+
h
x
_
t
is elliptic, since
1
= =
_
_
, det
1
=
2
2
1.8.2 Operators of hyperbolic type
We consider the product space V = X R and denote the coordinates
by x and t respectively. We have then V
= X
; the corresponding
coordinates are denoted by and . Write the principal symbol of an operator
a (x, t; D
x
, D
t
) in the form
m
(x, t; , ) = a
m
(x, t; , ) = (x, t) [
1
(x, t; )] ... [
m
(x, t; )]
Denition. We assume that in a domain D V
(i) (x, t) ,= 0, i.e. the time direction dt is not characteristic,
(ii) the roots
1
, ...,
m
are real for all X
,
(iii) we have
1
< ... <
m
for X
0 .
9
The operator a is called strictly t-hyperbolic ( strictly hyperbolic in vari-
able t), if (i,ii,iii) are fullled. It is called weakly hyperbolic, if (i) and (ii) are
satised. It is called t-hyperbolic, if there exists a number
0
< 0 such that
(x, t; + ) ,= 0, for V
, <
0
The strict hyperbolicity property implies hyperbolicity which, in its turn,
implies weak hyperbolicity. Any of these properties implies the same property
for t instead of t.
Example 1. The operator
=
2
t
2
v
2
2
=
2
+ v
2
(x) [[
2
= [ v (x) [[] [ + v (x) [[]
i.e.
1
= v [[ ,
2
= v [[ . It is strictly hyperbolic, if v (x) > 0.
Example 2. The Klein-Gordon-Fock operator + m
2
is strictly t-
hyperbolic.
Example 3. The Maxwell, Dirac systems are weakly hyperbolic, but
not strictly hyperbolic.
Example 4. The elasticity system is weakly hyperbolic, but not strictly
hyperbolic, since the polynomial det
2
is of degree 6 and has 4 real roots
with respect to , two of them of multiplicity 2.
1.8.3 Operators of parabolic type
Denition. An operator a (x, t; D
x
, D
t
) is called t-parabolic in a domain
U X R if the symbol has the form = (x, t) (
1
) ... (
p
) where
,= 0, and the roots full the condition
(iv) Im
j
(x, t; ) b [[
q
c for some positive constants q, b, c.
This implies that p < m.
Examples. The heat operator and the diusion operator are parabolic.
For the heat operator we have = + d
2
(x, t) [[
2
. It follows that p = 1,
1
= d
2
[[
2
and (iv) is fullled for q = 2.
10
1.8.4 Out of classication
The linear Schr odinger operator does not belong to either of the above classes.
Tricomi operator
a (x, y, D) =
2
u
x
2
+ x
2
u
y
2
is elliptic in the halfplane x > 0 and is strictly hyperbolic in x < 0 .
It does not belong to either class in the axes x = 0 .
1.9 Initial and boundary value problems
1.9.1 Boundary value problems for elliptic equations.
For a second order elliptic equation
a (x, D) u = f
in a domain D X the boundary conditions are: the Dirichlet condition:
u[D = v
0
or the Neumann condition:
u
[D = v
1
or the mixed (Robin) condition:
_
u
+ bu
_
[D = v
1.9.2 The Cauchy problem
u(x, 0) = u
0
for a diusion equation
a (x, t; D) u = f
For a second order equation the Cauchy conditions are
u(x, 0) = u
0
,
t
u(x, 0) = u
1
11
1.9.3 Goursat problem
u(x, 0) = u
0
, u(0, t) = v (t)
1.10 Inverse problems
To determine some coecients of an equation from boundary measurements
Examples
1. The sound speed v to be determined from scattering data of the acoustic
equation.
2. The potential V in the Schr odinger equation
3. The conductivity in the Poisson equation
and so on.
Bibliography
[1] R.Courant D.Hilbert: Methods of Mathematical Physics,
[2] P.A.M.Dirac: General Theory of Relativity, Wiley-Interscience Publ.,
1975
[3] L.Landau, E.Lifshitz: The classical theory of elds, Pergamon, 1985
[4] I.Rubinstein, L.Rubinstein: Partial dierential equations in classical
mathematical physics, 1993
[5] L.H.Ryder: Quantum Field Theory, Cambridge Univ. Press, London
1985
[6] V.S.Vladimirov: Equations of mathematical physics, 1981
[7] G.B.Whitham: Linear and nonlinear waves, Wiley-Interscience Publ.,
1974
12
Chapter 2
Elementary methods
2.1 Change of variables
Let V be an Euclidean space of dimension n with a coordinate system
x
1
, ..., x
n
. If we introduce another coordinate system, say y
1
, ..., y
n
, then we
have the system of equations
dy
j
=
y
j
x
1
dx
1
+ ... +
y
j
x
n
dx
n
, j = 1, ..., n
If we write the covector dx = (dx
1
, ..., dx
n
) as a column, this system can be
written in the compact form
dy = Jdx
where J
.
= y
j
/x
i
is the Jacobi matrix. For the rows of derivatives
D
x
= (/x
1
, ..., /x
n
) , D
y
= (/y
1
, ..., /y
n
) we have
D
x
= D
y
J
since the covector dx is bidual to the vector D
x
. Therefore for an arbitrary
linear dierential operator a we have
a (x, D
x
) = a (x(y) , D
y
J)
hence the symbol of a in y coordinates is equal (x(y) , J) , where (x, )
is symbol in x-coordinates.
Example 1. An arbitrary operator with constant coecients is invariant
with respect to arbitrary translation transformation T
h
: x x + h, h V.
Translations form the group that is isomorphic to V.
1
Example 2. DAlembert operator
=
2
t
2
v
2
2
x
2
, =
2
v
2
2
with constant speed v can be written in the form
= 4v
2
2
yz
where y = x vt, z = x + vt, since 2/y = /x v
1
/t, 2/z =
/x + v
1
/t.
This implies that an arbitrary solution u C
2
of the equation
u = 0
can be represented, at least, locally in the form
u(x, t) = f (x vt) + g (x + vt) (2.1)
for continuous functions f, g. At the other hand, if f, g are arbitrary con-
tinuous functions, the sum (1) need not to be a C
2
-function. Then u is a
generalized solution of the wave equation.
Example 3. The Laplace operator keeps its form under arbitrary
linear orthogonal transformation y = Lx. We have J = L and () = [[
2
.
Therefore () = [L[
2
= [[
2
. All the orthogonal transformations L
form a group O(n) . Also the Helmholtz equation is invariant with respect
to O(n) .
Example 4. The relativistic wave operator
=
2
t
2
c
2
= t cosh + c
1
x
j
sinh , x
j
= ct sinh + x
j
cosh , j = 1, ...
Dimension d of the Lorentz group is equal to n(n + 1) /2, in particular, d =
6 for the space dimension n = 3. The group generated by all translations
and Lorentz transformations is called Poincare group. The dimension fo the
Poincare group is equal 10.
2.2 Bilinear integrals
Suppose that V is an Euclidean space, dimV = n. The volume form dV
.
=
dx
1
... dx
n
is uniquely dened; let L
2
(V ) be the space of square integrable
functions in V. For a dierential operator a we consider the integral form
a, ) =
_
V
(x) a (x, D) (x) dV
It is linear with respect to the argument and is additive with respect to
whereas
a, ) = a, )
for arbitrary complex constant . A form with such properties is called
sesquilinear. It is bilinear with respect to multiplication by real constants.
We suppose that the arguments , are smooth (i.e. , C
) funtcions
with compact supports. We can integrate this form by parts up to m times,
where m is the order of a. The boundary terms vanish, since of the assump-
tion, and we come to the equation
a, ) = , a
) (2.2)
where a
=
a
, obviously a
= a.
An operator a is called (formally) selfadjoint if a
= a.
Example 1. For an arbitrary operator a with constant coecients we
have a
(D) = a (D) .
Example 2. A tangent eld b =
b
i
(x) /x
i
is a dierential operator
of order 1. We have
b
= b div b, div b
.
=
b
i
/x
i
3
This is no more a tangent eld unless the divergence vanishes.
Example 3. The Poisson operator
a (x, D) =
/x
i
(a
ij
(x) /x
j
)
is selfadjoint if the matrix a
ij
is Hermitian. In particular, the Laplace
operator is selfadjoint. Moreover the quadratic Hermitian form
, ) =
x
i
,
x
i
_
= ||
2
is always nonnegative. This property helps, f.e. to solve the Dirichlet problem
in a bounded domain. Note that the symbol of is also nonnegative:
[[
2
0. In general these two properties are related in much more general
operators.
Let a, b be arbitrary functions in a domain D V that are smooth up to
the boundary = D. They need not to vanish in . Then the integration
by parts brings boundary terms to the righthand side of (2). In particular,
for the Laplace operator we get the equation
_
D
badV =
_
D
a
x
i
b
x
i
dV +
_
n
i
a
x
i
dS
where n =(n
1
, ..., n
n
) is the unit outward normal eld in and dS is the
Euclidean surface measure. The sum of the terms n
i
a/x
i
is equal to the
normal derivative a/n. Integrating by parts in the rst term, we get nally
_
D
badV =
_
D
badV +
_
ba/ndS
_
b/nadS (2.3)
This is a Green formula.
2.3 Conservation laws
For some hyperbolic equations and system one can prove that the energy
is conservated, i.e. it does not depend on time. Consider for simplicity the
selfadjoint wave equation
_
2
t
2
x
i
v
2
x
i
_
u(x, t) = 0
4
in a space-time V = X R. Suppose that a solution u decreases as [x[
for any xed t and u stays bounded. Then we can integrate by parts in the
X-integral
_
x
i
v
2
_
x
i
u
_
, u
_
=
_
v
2
_
u
x
i
_
,
u
x
i
_
= |vu|
2
Take time derivative of the lefthand side:
t
_
2
u
t
2
, u
_
=
t
_
_
_
_
u
t
_
_
_
_
2
At the other hand from the equation
t
_
2
u
t
2
, u
_
=
t
_
x
i
_
v
2
u
x
i
_
, u
_
=
t
|vu|
2
and
t
_
_
_
_
_
u
t
_
_
_
_
2
+|vu|
2
_
= 0
Integrating this equation from 0 to t, we get the equation
_
_
_
_
_
_
u(x, t)
t
_
_
_
_
2
+|vu(x, t)|
2
_
dx =
_
_
_
_
_
_
u(x, 0)
t
_
_
_
_
2
+|vu(x, 0)|
2
_
dx
The left side has the meaning is the energy of the wave u at the time t.
2.4 Method of plane waves
Let again V be a real vector space of dimension n < and be a nonzero
linear functional on V. A function u in V is called a -plane-wave, if u(x) =
f ((x)) for a function f : R C. The function f is called the prole of
u. The meaning of the term is that any u is constant on each hyperplane
= const .
For example both the terms in (1) are plane waves for the covectors
= (1, v) and = (1, v) respectively. In general, if we look for a plane-
wave solution of a partial dierential equation, we get an ordinary dierential
equation for its prole.
5
Example 1. For an arbitrary linear equation with constant coecients
a (D) u = 0
the exponential function exp (x) is a solution if and only if the covector
satises the characteristic equation () = 0.
Example 2. For the Korteweg & de-Vries equation
u
t
+ 6uu
x
+ u
xxx
= 0
in R R and arbitrary a > 0 there exists a plane-wave solution for the
covector = (1, a):
u(x, t) =
a
2 cosh
2
(2
1
a
1/2
(x at x
0
))
It decreases fast out of the line x at =
0
. A solution of this kind is called
soliton.
Example 3. Consider the Liouville equation
u
tt
u
xx
= g exp (u)
where g is a constant. For any a, 0 a < 1 there exists a plane-wave solution
u(x, t) = ln
a
2
(1 a
2
)
2g cosh (2
1
a (x at x
0
))
Example 4. For the Sine-Gordon equation
u
tt
u
xx
= g
2
sin u
the function
u(x, t) = 4 arctan
_
exp
_
g
_
1 a
2
_
1/2
(x at x
0
)
__
is a plane-wave solution.
Example 5. The Burgers equation
u
t
+ uu
x
= u
xx
, ,= 0
It has the following solution for arbitrary c
1
, c
2
u = c
1
+
c
2
c
1
1 + exp
_
(2)
1
(c
2
c
1
) (x at)
_, 2a = c
1
+ c
2
6
2.5 Fourier transform
Consider ordinary linear equation with constant coecients
a (D) u =
_
a
m
d
m
dx
m
+ a
m1
d
m1
dx
m1
+ ... + a
0
_
u = w (2.4)
To solve this equation, we asume that w L
2
and write it by means of the
Fourier integral
w(x) =
_
exp (x) w() d
and try to solve the equation (4) for w(x) = exp (x) for any . Write a
solution in the form u
w()
()
exp (x) d
Example 1. The symbol of the ordinary operator a
= D
2
k
2
is equal
to =
2
k
2
,= 0. It does not vanish.
Proposition 1 If m > 0 and w has compact support, we can nd a solution
of (4) in the form
u(x) =
1
2
_
w()
()
exp (x) d (2.5)
where C = 0 is a cycle that is homologically equivalent to R in C.
Proof. The function w() has the unique analytic continuation w() at
C according to Paley-Wiener theorem. The integral (4) converges at innity,
since coincides with R in the complement to a disc, the function w()
belongs to L
2
and [ ()[ c [[ for [[ > A for suciently big A.
Example 2. The symbol of the Helmholtz operator a
+
= D
2
+ k
2
vanishes for = k. Take
+
C
+
= Im 0 . Then the solution (4)
vanishes at any ray x > x
0
, where w vanishes. If we take =
,
these rays will be replaced by x < x
0
.
7
2.6 Theory of distributions
See Lecture notes FI3.
Bibliography
[1] R.Courant D.Hilbert: Methods of Mathematical Physics
[2] I.Rubinstein, L.Rubinstein: Partial dierential equations in classical
mathematical physics, 1993
[3] V.S.Vladimirov: Equations of mathematical physics, 1981
[4] G.B.Whitham: Linear and nonlinear waves, Wiley-Interscience Publ.,
1974
8
Chapter 3
Fundamental solutions
3.1 Basic denition and properties
Denition. Let a (x, D) be a linear partial dierential operator in a vec-
tor space V and U is an open subset of V. A family of distributions F
y
D
) dx
= E
y
(x) dx, where x
= x
(x) , hence E
y
(x
) = E
y
(x) [det x/x
[ .
The function E
y
for a xed y is called a source function with the source point
y.
If a (D) is an operator with constant coecients in V and E
0
(x) = E (x)
is a source function that satises a (D) E
0
=
0
, then E
y
(x) dx
.
= E (x y) dx
is a f.s in U = V. Later on we call E source function; we shall use the same
notation E
y
for a f.s. and corresponding source function, if we do not expect
a confusion.
Proposition 1 Let E be a f.s. for U V. If w is a function (or a distribution)
with compact support K U, then the function (distribution)
u(x)
.
=
_
E
y
(x) w(y) dy
is a solution of the equation a (x, D) u = w.
1
Proof for functions E and w
a (x, D) u =
_
a (x, D) E
y
(x) w(y) dy =
_
y
(x) w(y) dy = w(x)
The same arguments for distributions E, w and u:
a (x, D) u() = u(a
(x, D) ) = w(F
y
(a
(x, D) )) = w()
0
w = w.
Reminder. For arbitrary distribution f and a distribution g with compact
support in V the convolution is the distribution
f g () = f g ((x + y)) , D (V )
Here f g is a distribution in the space V
x
V
y
,where both factors are isomorphic
to V , x and y are corresponding coordinates.
If the order of a is positive, there are many fundamental solutions. If E is
a f.s. and U fulls a (D) U = 0, then E
.
= E + U is a f.s. too.
Theorem 2 An arbitrary dierential operator a ,= 0 with constant coecients
possesses a f.s.
Problem. Prove this theorem. Hint: modify the method of Sec.4.
3.2 Fundamental solutions for elliptic opera-
tors
Denition. A linear dierential operator a (x, D) is called elliptic in an open
set W V, if the principal symbol
m
(x, ) of a does not vanish for
V
0 , x W.
Now we construct fundamental solutions for some simple elliptic operators
with constant coecients.
Example 1. For the ordinary operator a (D) = D
2
k
2
we can nd a f.s.
by means of a formula (5) of Ch.2, where w =
0
and w = 1 :
E (x) =
1
2
_
R
exp (x)
2
+ k
2
d = (2k)
1
exp (k [x[)
Example 2. For the Helmholtz operator a (D) = D
2
+ k
2
we can nd a
f.s. by means of (5), Ch.2, where w =
0
and w = 1 :
E (x) =
1
2
_
exp (x)
k
2
2
d
2
If we take = 1/2 (
+
+
dxdy
z
= (0)
for any D (R
2
) , where a
=
z
. The kernel z
1
is locally integrable, hence
the integral is equal to the limit of integrals over the set U () = [z[ as
0. We have a
=
z
; integrating by parts yields
_
U()
z
dxdy
z
=
1
_
U()
z
_
1
z
_
dxdy (2)
1
_
U()
dy dx
z
The rst term vanishes since the function z
1
is analytic in U () . The bound-
ary U () is the circle [z[ = with opposite orientation. Take the parametriza-
tion by x = cos , y = sin and calculate the second term:
(2)
1
_
U()
dy dx
z
= (2)
1
_
2
0
( cos , sin ) d
(2)
1
(0)
_
2
0
d = (0)
Example 4. For the Laplace operator in the plane V = R
2
the function
E = (2)
1
ln [x[
is a f.s. To check it we apply the Green formula to the domain D = U () (see
Ch.2)
E () = lim
0
_
D
EdV = lim
0
__
D
EdV +
_
E/ndS
_
E/ndS
_
Here E = 0 in D, = U () , /n = /r, E/n = (2r)
1
, r = [x[
and
E () = (2)
1
_
_
r=
ln r /rdS +
_
r=
r
1
dS
_
The rst integral tends to zero as 0, since the function /r is bounded
and r ln r 0. In the second one we have r
1
dS = d, hence the integral
tends to 2(0) , which implies E () = (0) . This means that E =
0
,
Q.E.D.
The function E is invariant with respect to rotations. This is the only
rotation invariant f.s. up to a constant term.
3
Example 5. The function E (x) = (4 [x[)
1
is a f.s. for the Laplace
operator in R
3
.
Problem. To check this fact.
Here are the basic properties of elliptic operators:
Theorem 3 Let a (x, D) be an elliptic operator with C
-coecients in an
open set W V. An arbitrary distribution solution to the equation
a (x, D) u = w
in W is a C
-
function in the complement to the source point.
Example 6. For DAlembert operator a (D) =
2
.
=
2
t
v
2
2
x
we can
nd a fundamental solution by means of the coordinate change y = x vt,
z = x + vt (see Ch.2):
2
= 4v
2
z
. Introduce the function (Heavisides
function)
(x) = 1, for x 0, (x) = 0 for x < 0
We have
x
(x) =
0
, hence we can take
E (y, z) =
_
4v
2
_
1
(y) (z) (3.1)
Returning to the space-time coordinate we get
E (x, t) = (2v)
1
(vt [x[)
i.e. E (x, t) = (2v)
1
if vt [x[ and E (x, t) = 0 otherwise. The coecient 2v
appears because of E is a density, (or a distribution), not a function.
We can replace (y) to (y) and (z) to (z) in (1) and get three
more options for a f.s.
Example 7. Consider the rst order operator a (D) =
a
j
j
, with
constant coecients a
j
R and introduce the variables y
1
, ..., y
n
such that
a (D) y
1
= 1, a (D) y
j
= 0, j = 2, ..., n and det y/x = 1. Then the general-
ized function E (x) = (y
1
)
0
(y
2
, ..., y
n
) is a fundamental solution for a.
Problem. To check this statement.
4
Forward propagator in 2spacetime
vt> |x|
=
E=1/2v
3.4 Hyperbolic polynomials and source func-
tions
Denition. Let p () be a polynomial in V
with complex coecients. It is
called hyperbolic with respect to a vector V
0 , if there exists a number
Let p
m
be the principal part of p. It is called strictly hyperbolic, if the equation
() = p
m
( + ) = 0 has only real zeros for real and these zeros are simple
for ,= 0. If p
m
is strictly hyperbolic, then p is hyperbolic for arbitrary lower
order terms.
Denition. Let a (x, D) be a dierential operator in V and t be a linear
function in V, called the time variable. The operator a is called hyperbolic with
respect to t (or t-hyperbolic) in U V , if the symbol (x, ) is a hyperbolic
polynomial in with respect to the covector (x) = t for any point x U.
Let p be a hyperbolic polynomial with respect to . Consider the cone
V
p
m
() = 0 and take the connected component (p, ) of this cone that
contains . This is a convex cone and p is hyperbolic with respect to any
g (p, ) and g (p, ) . The dual cone is dened as follows
(p,
0
)
5
where p is the principal symbol of a.
Proof. Fix a covector (a,
0
) , [[ = 1 and set
E (x) = (2)
n
lim
0
E
,
(x) (3.2)
E
,
(x) =
_
V
exp ( + , x)) exp
_
( + )
2
_
p ( + )
d
where n = dimV and p is the symbol of a. The dominator p ( + ) does
not vanish as V
, (p,
0
) , <
, since a is
0
-hyperbolic. The
integral converges at innity since of the decreasing factor exp (
2
) and
commutes with any partial derivative. The integrand can be extended to C
n
as a meromorphic dierential form
= p
1
() exp
_
x
2
_
d
that is holomorphic in the cone V
+ (a,
0
) C
n
. The integral of does
not depend on in virtue of the Cauchy-Poincare theorem (the special case of
the general Stokes theorem). Check that it is a fundamental solution:
a (D) E
,
(x, t) =
m
_
V
p ( + ) exp ( + , x)) exp
_
( + )
2
_
p ( + )
d
=
_
V
exp ( + , x)) exp
_
( + )
2
_
d
=
_
V
exp (x) exp
_
2
_
d
= ()
n/2
exp
_
(4)
1
_
[x[
2
__
(2)
n
0
in S
For the third step we have applied again the Cauchy-Poincare theorem. This
yields a (D) E =
0
. Estimate this integral in the halfspace , x) < 0 :
[E
(x)[ =
_
V
exp ( , x)) exp ( , x)) exp
_
( + )
2
_
p ( + )
d
exp ( , x))
_
V
exp (
2
) exp (
2
2
)
[p ( + )[
d C
n/2
exp
_
2
_
exp ( , x))
for a constant C, since of [p ( + )[ c
0
> 0. We take =
2
;the right side
is then equal to O(
n
exp ( , x))) . This quantity tends to zero as ,
since , x) < 0. Therefore E (x) = 0 in the half-space H
.
= , x) < 0 .
Therefore E vanishes in the union of all half-spaces H
, (p,
0
) . The
complement to this union in V is just
(p,
0
) . This completes the proof.
Proposition 6 Let W V be a closed half-space such that the conormal is
not a zero the symbol a
m
. If u a solution to the equation a (D) u = 0 supported
by W, then u = 0.
6
Proof. Take a hyperplane H V W. The distribution U vanishes in a
neighborhood of H and by Holmgren uniqueness theorem (see Ch.4) it vanishes
everywhere.
Corollary 7 If a a hyperbolic operator with constant coecients with respect
to a covector , then there exists only one fundamental solution supported by
the set , x) 0 .
3.5 Wave propagators
The wave operator
n
.
=
2
t
2
v
2
x
in V = X R with a positive velocity v = v (x) is hyperbolic with respect to
the time variable t. The symbol (, ) =
2
+ v
2
[[
2
is strictly hyperbolic
with respect to the covector
0
= (0, 1) . Now we assume that the velocity v
is constant; the wave operator is hyperbolic and with respect to any covector
(, 1) such that v [[ < 1 and the union of these covectors is just the cone
(,
0
) . The dual cone is
(,
0
) = (x, t) , v [x[ t
The f.s. supported by this cone is called the forward propagator. For the
opposite cone
(x, t) (3.4)
E
(x, t) =
_
X
exp (x + ( ) t) exp (
2
)
( )
2
[v[
2
dd
7
where , are coordinates dual to x, t, respectively and we write x instead
of , x) . The interior integral converges, hence we need not the decreasing
factor exp (
2
) . We know that E vanishes for t < 0; assume that t > 0. The
backward propagator vanishes, hence we can take the dierence as follows:
E
(x, t) =
_
X
exp
_
2
_
d
_
_
exp (x + ( ) t)
( )
2
[v[
2
exp (x + ( + ) t)
( + )
2
[v[
2
_
d
The interior integral is equal to the integral of the meromorphic form =
_
2
[v[
2
_
1
exp ( , x) + t) over the chain Im = Im = , which
is equivalent to the union of circles [ [v[[ = [ +[v[[ = . By the
residue theorem we nd
_
...d = 2 exp (x)
exp (vt [[) exp (vt [[)
2v [[
= 2 exp (x)
sin (vt [[)
v [[
consequently
E
(x, t) = 2v
1
_
X
exp (x)
sin (vt [[)
[[
exp
_
2
_
d
= 2v
1
F
_
sin (vt [[)
[[
exp
_
2
_
_
Lemma 9 We have for an arbitrary a > 0
F
_
S(a)
_
= 4a
sin (a [[)
[[
(3.5)
where
S(a)
denotes the delta-density on the sphere S (a) of radius a :
S(a)
() =
_
S(a)
dS
Proof of Lemma. For an arbitrary test function D(X
) we have
F
_
S(a)
_
() =
S(a)
(F ()) =
S(a)
__
exp (x) () d
_
=
_
()
S(a)
(exp (x)) d
The functional
S(a)
has compact support and therefore is well dened on the
smooth function exp (x) . Calculate the value:
S(a)
(exp (x)) =
_
S(a)
exp (x) dS = a
2
_
2
0
_
0
exp (a [[ cos ) sin dd
= 2a
2
exp (a [[) exp (a [[)
a [[
= 4a
sin (a [[)
[[
8
This implies (5).
We have F
F = (2)
3
I, where I stands for the identity operator, hence
by (4)
F
_
sin (a [[)
[[
_
= 2
2
a
1
S(a)
We nd from (5)
E
(x, t) = 2v
1
F
_
sin (vt [[)
[[
exp
_
2
_
_
4
3
v
2
t
1
S(vt)
This implies (3) in virtue of (4).
Problem. Calculate the backward propagator for
4
.
Note that the support of E
4
is the conic 3-surface S = vt = [x[ , see the
picture
Proposition 10 For n = 3 the forward propagator is equal to
E
3
(x, t) =
(vt [x[)
2v
_
v
2
t
2
[x[
2
(3.6)
Replacing (vt [x[) by (vt [x[) , we obtain the backward propagator.
The support of the convex cone vt [x[ in E
3
. The prole of the function
E
3
is shown in the picture
9
Profile of 3spacetime propagator
x=vt
E=c((vt) x )
2 2 1/2
Proof. We apply the dimension descent method. Write x = (x
1
, x
2
, x
3
)
in (3) and integrate this function for xed y = (x
1
, x
2
) against the density dx
3
from to . The line (x
1
, x
2
) = y meets the surface S only if t v
1
[y[ .
Therefore the function
E
3
(y, t)
.
=
_
E
4
(x, t) dx
3
= (2v)
1
_
(vt [x[) dx
3
is supported by the cone K
3
.
= vt [y[ . Apply this equation to a test
function:
E
3
() = E
4
( e) =
_
4v
2
t
_
1
(vt [x[) ( e) =
_
4v
2
t
_
1
_
y
2
+x
2
3
=(vt)
2
(y, t) dS
where e = e (x
3
) = 1. Consider the projection p : R
3
R
2
, x y = (x
1
, x
2
) .
The mapping p : S K covers the cone K twice and we have n
3
dS = dx
1
dx
2
,
where n is the normal unit eld to S and n
3
= (vt)
1
_
(vt)
2
[y[
2
_
1/2
. It
follows
dS =
vtdx
1
dx
2
_
v
2
t
2
[y[
2
and
E
3
() =
1
2v
_
(y) dx
1
dx
2
_
v
2
t
2
[y[
2
which coincides with (6). We need only to check that E
3
is the forward prop-
agator for the operator
3
. It is supported by the proper convex cone K and
3
E
3
=
_
4
E
4
dx
3
=
_
0
(x, t) dx
3
=
0
(y, t)
since
_
2
3
E
4
dx
3
= 0.
10
3.6 Inhomogeneous hyperbolic operators
Example 7. The forward propagator for the Klein-Gordon-Fock operator
4
+ m
2
is equal to
D(x, t) =
(t)
4c
2
t
(ct [x[)
m
4
(ct [x[)
J
1
_
m
_
c
2
t
2
[x[
2
_
_
c
2
t
2
[x[
2
(3.7)
where J
1
is a Bessel function. Recall that the Bessel function of order can
be given by the formula
J
(z) =
0
(1)
k
k!( + k + 1)
_
z
2
_
+2k
Remark. The generalized functions in (3), (6), and (7) can be written as
pullbacks of some functions under the mapping
X R R
2
, (x, t) q = v
2
t
2
[x[
2
, (t) (3.8)
It is obvious for (6) since (vt [x[) = (q) (t) . Fix the coordinates (x
0
, x)
in V, where x
0
= vt. In formulae (3) and (7) we can write (ct)
1
(ct [x[) =
(q) . Indeed, we have by denition
(q) () =
_
q=0
dq
=
1 + v
2
_
0
_
q=0
[q[
dSdt
where = dxdx
0
is a test density, i.e. D(X R); dS is the area
in the 2-surface q = 0, t = const . We have q = (2x, 2v
2
t) = (2x, 2v
2
t) ,
[q[ = 2
1 + v
2
vt and
_
q=0
[q[
dS =
_
2
1 + v
2
vt
_
1
_
dS
Then
E
4
=
(t)
4v
2
t
([x[ vt) =
(t)
2v
(q)
D = (t)
_
1
2c
(q)
m
4
(q)
J
1
_
m
q
_
q
_
This fact has the following explanation. The wave operator and the Klein-
Gordon-Fock operator are invariant with respect to the Lorentz group L
3
=
O(3, 1) . This is the group of linear transformations in V that preserve the
quadratic form q; the dual transformations in V
preserve the dual form
11
q
(, ) =
2
c
2
[[
2
. Any Lorentz transformation preserves the volume form
dV = dxdt too. Therefore the variety of all source functions is invariant with
respect to this group. The forward propagator is uniquely dene. Therefore it
is invariant with respect to the orthochronic Lorentz group L
3
, i.e. to group of
transformations A L
3
that preserves the time direction. The functions q and
sgn t are invariant of the orthochronic Lorentz group and any other invariant
function (even a generalized function) is a function of these two. We see that
is the fact for the forward propagators (as well as for backward propagators).
Example 8. The function
D
c
(x, t)
.
= (2)
4
_
X
_
R
exp (t + (, x)) dd
2
+
is also a fundamental solution for the wave operator
4
. The integral must
be regularized at innity by introducing a factor like exp (
2
) ; it does not
depend on > 0 since the dominator has no zeros in V
= X
. This
function is called causal propagator and plays fundamental role in the tech-
niques of Feynman diagrams ? It is invariant with respect to the complete
Lorentz group L
3
. The causal propagator vanishes in no open set, hence it is
not equal to a linear combination of the forward and backward propagators.
The causal propagator for the Klein-Gordon-Fock operator is dened in by the
same formula with the extra term m
2
in the dominator.
3.7 Riesz groups
This construction provides an elegant and uniform method for explicit con-
struction of forward propagators for powers of the wave operator in arbitrary
space dimension.
Let V be a space of dimension n with the coordinates (x
1
, ..., x
n
); set q (x) =
x
2
1
x
2
2
... x
2
n
. The set K
.
= x
1
0, q (x) 0 is a proper convex cone in
V. Consider the family of distributions
q
+
() =
_
K
q
+
() is an analytic function of for any . The family has a meromorphic
continuation to whole plane C with poles at the points
= 0, 1, 2, ...; =
n
2
1,
n
2
2, ...
and after normalization
Z
.
=
q
n/2
+
dx
(n2)/2
2
21
() ( + 1 n/2)
,
x
1
+
dx
()
(3.9)
12
becomes an entire function of with values in the space of tempered distri-
butions. We have always supp Z
K, hence Z
= Z
+
(3.10)
The points = 0, 1, 2, ... are poles of the numerator and denominator in
(9) and the value of Z
is a
commutative group, which is isomorphic to the additive group of C. It is called
the Riesz group. From (11) we see that
k
Z
k
= Z
k
Z
k
= Z
0
=
0
dx
This means that Z
k
is a fundamental solution for the hyperbolic operator
k
(which is not strictly hyperbolic for k > 1). Moreover it is a forward
propagator, since supp Z
k
K.
If the dimension n is even, the point = k = n/2 1 is again a pole of the
numerator and denominator in (9), as a consequence of which the support of
Z
k
is contained in the boundary K. This fact is an expression of the strong
Huyghens principle: for even dimension the wave initiated by a local source
has back front, whereas the forward front exists for arbitrary dimension. This
is just the case for n = 4, k = 1.
References
[1] L.Schwartz: Theori`e des distributions (Theory of distributions)
[2] R.Courant, D.Hilbert: Methods of Mathematical Physics
[3] I.Rubinstein, L.Rubinstein: Partial dierential equations in classical
mathematical physics
[4] F.Treves: Basic linear partial dierential equations
[5] V.S.Vladimirov: Equations of mathematical physics
13
Chapter 4
The Cauchy problem
4.1 Denitions
Let a (x, D) be a linear dierential operator of order m with smooth coecients in the
space V
n
and W be an open set in V. Let t be a smooth function in W such that
dt = 0 (called time variable) and f, g be some functions in W. The Cauchy problem for
time variable t for the data f, g is to nd a solution u to the equation
a (x, D) u = f (4.1)
in W that fulls the initial condition
u g = O(t
m
)
in a neighborhood of W
0
.
= {x W, t (x) = 0} .
First, assume that the right-hand side f and g are smooth. Introduce the coordinates
x
= (x
1
, ..., x
n1
) (space variables) such that (x
m
t
u +
1
m1
t
u +... +
m
u = f (4.2)
where
j
, j = 0, 1, ..., m is a dierential operator of order j which does not contain
time derivatives. In particular,
0
is a function. The initial condition can be written in
the form
u|
t=0
= g
0
,
t
u|
t=0
= g
1
, ...,
m1
t
u|
t=0
= g
m1
where g
j
=
j
t
g|
t=0
, j = 0, ..., m1 are known functions in W
0
. Set t = 0 in (2) and nd
m
t
u|
t=0
=
f
1
m1
t
u ...
m
u
|
t=0
= f|
t=0
1
g
m1
...
m
g
0
from this equation we can nd the function
m
t
u|W
0
, if
0
|W
0
= 0. Take t-derivative of
both sides of (2) and apply the above arguments to determine
m+1
t
u|W
0
and so on.
Denition. The hypersurface W
0
is called non-characteristic for the operator a at
a point x W
0
, if
0
(x) = 0. Note that
0
(x) =
m
(x, dt (x)) , where
m
|W V
is the
principal symbol of a and V
, (x) = t. An arbitrary smooth hypersurface H V
1
is non-characteristic at a point x, if
m
(x, ) = 0, where is the conormal vector to H
at x.
The necessary condition for solvability of the Cauchy for arbitrary data is that the
hypersurface W
0
is everywhere non-characteristic. This condition is not sucient. For el-
liptic operator a an arbitrary hypersurface is non-characteristic, but the Cauchy problem
can be solved only for a narrow class of initial functions g
0
, ..., g
m1
.
Example 1. For the equation
2
u
tx
= 0
the variable t as well as x is characteristic, n = 2;
2
= . dt = (0, 1) ;
2
(0, 1) = 0.
Example 2. For the heat equation
t
u
x
u = 0
the variable t is characteristic, but the space variables are not. u|t = 0 = u
0
.
Example 3. The Poisson equation u = 0 is elliptic, but the Cauchy problem
u|W
0
= g
0
,
t
u|W
0
= g
1
has no solution in W, unless g
0
and g
1
are analytic functions. In fact, it has no solution
in the half-space W
+
, if g
0
, g
1
are in L
2
(W
0
), unless these functions satisfy a strong
consistency condition.
4.2 Cauchy problem for distributions
The non-characteristic Cauchy problem can be applied to generalized functions as well.
First, we write our space as the direct product V = XR by means of coordinates x
and
t. For arbitrary test densities and in X and R, respectively, we can take the product
(x
, t) = (x
() = u() =
, v
Denition. The function u is called weakly smooth in t-variable (or t-smooth), if the
functional u
j
t
E|
t=
j
m1
m
()
0
(x) as 0, j = 0, ..., m1 (4.3)
3
Proof of Lemma. Apply the formula (2) of Ch.3
E (x, t) = (2)
n
lim
0
E
,
(x, t)
E
,
(x, t) =
exp (( + ) t)
a (, + )
d exp
x ||
2
d, <
exp (t) d
a (, )
where = {Re = } . All the zeros of the dominator are to the left of . By Cauchy
Theorem we can replace by a big circle
m1
t
E
,
(x, t) 2
1
0
1
0
0
(x)
Taking in account that
0
= a
m
() , we complete the proof.
Note that for any higher derivative
j
t
E the limit as (4) exists and can be found from
(4) and the equation a (D) E = 0 for t > 0.
Proof of Theorem. First we dene a solution u of (1) by
u = E f
The convolution is well dened, since supp f H
+
and supp E K, the cone K is
convex and proper. The distribution u is weakly smooth in t, hence the initial data of it
are well dened. Therefore we need now to solve the Cauchy problem for the equation
a (D) u = 0 (4.4)
with the initial conditions
u|
t=0
= g
0
,
u|
t=0
= g
1
, ...,
m1
u|
t=0
= g
m1
(4.5)
where g
j
= u
j
j
t
u|
t=0
, j = 0, ..., m1. Take rst the convolution
e
0
.
=
0
m1
t
E g
0
(t, x) =
0
m1
t
E (t, x y) g
0
(y) dy (4.6)
This is a solution of (5); according to Lemma and e
0
|
t=0
= g
0
. The derivatives
j
t
e
0
|
t=0
can be calculated by dierentiating (7), since any time derivative of E has a limit as
4
t 0. Therefore we can replace the unknown function u by u
.
= u e
0
. The function u
m2
t
E g
1
By Lemma we have e
1
|
t=0
= 0 and
t
e
1
|
t=0
= g
1
. Then we replace u by u
= u
e
1
and
so on.
4.4 Solution of the Cauchy problem for wave equa-
tions
Applying the above Theorem to the wave equations with the velocity v, we get the
classical formulae:
Case n = 2. The DAlembert formula
2vu(x, t) =
t
0
x+v(ts)
xv(ts)
f (y, s) dyds
+
x+vt
xvt
g
1
(y) dy + v [g
0
(x + vt) + g
0
(x vt)]
Case n = 3. The Poisson formula
2vu(x, t) =
t
0
B(x,v(ts))
f (y, s) dyds
v
2
(t s)
2
|x y|
2
+
B(x,vt)
g
1
(y) dy
v
2
t
2
|x y|
2
+
t
B(x,vt)
g
0
(y) dy
v
2
t
2
|x y|
2
Case n = 4. The Kirchhof formula
4v
2
u(x, t) =
B(x,vt)
f (y, t v
1
|x y|) dy
|x y|
+
S(x,vt)
g
1
(y) dS +
t
t
1
S(x,vt)
g
0
(y) dS
Here B(x, r) denotes the ball with center x, radius r; S (x, r) is the boundary of this ball.
4.5 Domain of dependence
Assume for simplicity that the right side vanishes: f = 0. The solution in a point (x, t)
does not depend on the initial data out of the ball B(x, vt) , i.e. a wave that is initiated
by the initial functions g
0
and g
1
is propagated with the nite velocity v. This is called
the general Huygens principle. In the case n = 3 the wave propagating from a compact
source has back front (see the picture). This is called the special Huygens principle
(Minor premiss).
5
0
10
20
30
40
50
60
70
0
20
40
60
80
2.5
2
1.5
1
0.5
0
0.5
1
1.5
2
Domain of dependence
(x,t)
W
Domain of dependence in 4D space
(x,t)
6
The special Huygens principle holds for the wave equation with constant velocity in
the space of arbitrary even dimension n 4.
References
[1] R.Courant D.Hilbert: Methods of mathematical physics
[2] F.Treves: Basic linear partial dierential equations
[3] V.S.Vladimirov: Equations of mathematical physics
7
Chapter 5
Helmholtz equation and
scattering
5.1 Time-harmonic waves
Let a (x, D
x
, D
t
) be a linear dierential operator of order m with smooth coe-
cients in the space-time V = X
n
R with coordinates (x, t) , whose coecients
do not depend on the time variable t. Consider the equation
a (x, D
x
, D
t
) U (x, t) = F (x, t) (5.1)
A function of the form F (x, t) = exp (t) f (x) is called time-harmonic of fre-
quency , the function f is called the amplitude. If a solution is also time-
harmonic function U (x, t) = exp (t) u(x), we obtain the time-independent
equation for the amplitudes
a (x, D
x
, ) u(x) = f (x)
Example 1. For the Laplace operator in space-time a (D
x
, D
t
) = D
2
t
+
X
we
have
a (D
x
, ) =
2
+
This is a negative operator. Therefore any solutions of the equation (
2
) u =
0 in X of nite energy i.e. u L
2
(X) decreases fast at innity.
Example 2. If a = D
2
t
v
2
(x) is the wave operator and the velocity v does
not depend on time, then
a (x, D
x
, ) =
2
+v
2
(x)
is the Helmholtz operator. The Helmholtz equation with f = 0 is usually written
in the form
_
+n
2
2
_
u = 0
where the function n
.
= v
1
is called the refraction coecient. The Helmholtz
operator is not denite; there are many oscillating bounded solutions of the form
u(x) = exp (x) , where
.
= n
2
2
= 0, R
n
. This solution is unbounded,
when C
n
\R
n
.
Find a fundamental solution for the time-independent equation:
1
Proposition 1 Let E (x, t) be a fundamental solution for (1) that can be repre-
sented by means of the Fourier integral
E
y
(x, t) =
1
2
_
exp (t)
E
y
(x, ) d (5.2)
for a tempered (Schwartz) distribution
E. Then
a (x, D
x
, )
E
y
(x, ) =
y
(x) (5.3)
i.e.
E (x, ) is a source function for the operator a (x, D
x
, ) for any such that
E is weakly -smooth.
Proof. We have
y
(x, t) = a (x, D
x
, D
t
) E
y
(x, t) =
1
2
_
exp (t) a (x, D
x
, )
E
y
(X, ) d
At the other hand
0
(t) =
1
2
_
exp (t) d,
y
(x, t) = (t)
y
(x) =
1
2
_
exp (t) d
y
(x)
Comparing, we get
_
exp (t) a (x, D
x
, )
E (x, ) d =
_
exp (t) d (x)
which implies (3) in the sense of generalized functions in V R
. If
E
y
(x, ) is -
smooth for some
0
we can consider the restriction of both sides to the hyperplane
=
0
. Then we obtain (3).
5.2 Source functions for Helmholtz equation
Apply this method to the wave operator with a constant velocity v. Take the
forward propagator E. It is supported in {t n|x|} and bounded as t .
Therefore it can be represented by means of the Fourier integral (2) for the
tempered distribution
E (x, ) =
_
n|x|
E (x, t) exp (t) dt (5.4)
The corresponding source function for the Helmholtz operator is equal F
n
(x, )
.
=
v
2
E
n+1
(x, ) . Calculate it:
Case n = 1. We have E
2
(x, t) = (2v)
1
(vt |x|) and
F
1
(x, ) =
_
n|x|
exp (t) dt =
2n
exp (n|x|)
2
Case n = 3. We have
F
3
(x, ) =
1
4n|x|
_
0
(|x| vt) exp (t) dt =
exp (n|x|)
4 |x|
Case n = 2. We have
F
2
(x, ) =
1
2
_
n|x|
exp (t)
_
t
2
n
2
|x|
2
dt
This integral is not an elementary function; it is equal to c
0
H
(1)
0
(n|x|) , where
H
(1)
0
is a Hankel function. The equation F
2
(x) = (2)
1
ln |x| +R(x, ) , where
R is a C
1
-function in a neighbourhood of the origin.
Proposition 2 The function F
n
(x, ) is the boundary value at the ray { > 0}
of a function F
n
(x, ) that is holomorphic in the half-plane { = +, > 0} .
Proof. The integral (4) has holomorphic continuation at the opposite half-
plane.
5.3 Radiation condition
Let K be a compact set in X
3
with smooth boundary and connected complement
X\K. Consider the exterior boundary problem
_
+k
2
_
u(x) = 0, x X\K, k = n (5.5)
u(x) = f (x) , x K (Dirichlet condition)
where f is a function on the boundary. Any solution is a real analytic function
u = u(x) since the Helmholtz operator is of elliptic type. A solution is not
unique, unless an additional condition is imposed. The radiation (Sommerfeld)
condition is as follows
u
r
ku = o
_
r
1
_
as r = |x| ; u
r
= u/r (5.6)
Theorem 3 If k > 0 and f = 0, there is only trivial solution u = 0 to the
exterior problem satisfying the radiation condition.
Proof. Let S (R) denote the sphere {|x| = R} in X. We have S (R) X\K
for large R R
0
. Write for an arbitrary solution u :
_
S(R)
|u
r
ku|
2
dS =
_
S(R)
_
|u
r
|
2
+k
2
|u|
2
_
dS k
_
S(R)
(u
r
u uu
r
) dS (5.7)
By (6) the left side tends to zero as R . At the other hand, by Greens
formula
_
S(R)
(u
r
u uu
r
) dS =
_
K
(
uu u
u) dS
3
where
stands for the normal derivative on K. The right side vanishes, since
u|K = 0, hence (7) implies
_
S(R)
|u|
2
dS 0, R (5.8)
Lemma 4 [Rellich] For k > 0 any solution u of the Helmholtz equation in X\K
satisfying (8) equals identically zero.
Proof. Consider the integral
U (r)
.
=
_
S
2
u(rs) (s) ds
where is a continuous function and ds is the Euclidean area density in the unit
sphere S
2
. We prove that U (r) = 0 for r > R
0
and for each eigenfunction of
the spherical Laplace operator
S
:
S
=
R()
.
= (
S
Id)
1
. In view of the formula
=
2
r
+ 2r
1
r
+r
2
S
it follows that U (r) satises the ordinary equation
U
rr
+ 2r
1
U
r
+
_
k
2
+r
2
_
U = 0, r > R
0
This dierential equation has two solutions of the form
U
(r) = C
r
1
exp (kr) +o
_
r
1
_
, r
Clearly, no nontrivial linear combination of U
+
and U
is o (r
1
) . A the other
hand the hypothesis implies that U (r) = o (r
1
) ; we deduce that U = 0. The
operator
S
is self-adjoint non-positive and the resolvent is compact. The set of
eigenfunctions is a complete system in L
2
(S
2
) by Hilberts theorem. This implies
that u = 0 for |x| > R
0
. The function u is real analytic, consequently it vanishes
everywhere in X\K.
Now we state existence of a solution of the problem (5).
Theorem 5 [Kirchhof-Helmholtz] If the function f is suciently smooth, then
there exists a solution of (5) satisfying the radiation condition. This solution is
of the form
u(x) =
_
K
_
f (y)
exp (k |x y|)
|x y|
g (y)
exp (k |x y|)
|x y|
_
dS
y
, g =
u
(5.9)
4
Sketch of a proof. First we replace the Helmholtz operator by +
(k +)
2
. The function F
3
(x, k +)
.
= (4 |x|)
1
exp ((k ) |x|) is a funda-
mental solution which coincides with F
3
(x, k) for = 0. The symbol equals
= ||
2
+(k +)
2
and |
|
2
> 0. Therefore there exists a unique function
u
L
2
(X\K) that satises the conditions
_
+ (k +)
2
_
u
= 0 in X\K
u
|K = f
(This fact follows from standard estimates for solutions of a elliptic boundary
value problem.) Moreover the sequence u
(x) =
_
K
_
S(R)
_
f (y)
r
exp (k |x y|)
|x y|
=
_
x
|x|
,
_
exp (k |x y|)
|x y|
=
(x, x y)
|x| |x y|
k exp (k |x y|)
|x y|
+O
_
|x|
2
_
=
k exp (k |x y|)
|x y|
+O
_
|x|
2
_
The kernel in the rst term (double-layer) equals
sin
+ sin
2
.
Problem. Show that the operator
S
is self-adjoint non-positive and the
resolvent is compact.
Remark. The radiation condition is a method to single out a unique solution
of the exterior problem. The real part of this solution is physically relevant, in
particular,
F
3
=
cos (k |x|)
4 |x|
is also a source function. Therefore we can replace to simultaneously in (4),
(6) and (9).
5
5.4 Scattering on an obstacle
The plane wave u
i
(x) = exp (k (, x)) is a solution of the Helmholtz equation
in the free space X for arbitrary (incident) unit vector . Let K be a compact
set in X, called obstacle. It impose a boundary value condition to any solution.
There are several types of such conditions. We suppose that the obstacle is
impenetrable and the eld u satises the Dirichlet condition u|K = 0. In this
case the boundary K is called also soft or pressure release surface in the context
of the acoustic wave theory. In the case of Neumann condition
u|K = 0 it is
called hard surface, the third condition appears for impedance surface. The total
eld u = u
i
+ u
s
is the sum of the incident plane wave and the scattered eld
u
s
(; x) in X\K such that u satises the Dirichlet condition
u
s
|K = exp (k (, x)) |K
and u
s
fulls the radiation condition. According to the above theorem the scat-
tered eld exists and unique. Moreover, by (9) it can be represented in the form
u
s
(; x) =
exp (k |x|)
4 |x|
A
_
;
x
|x|
_
+O
_
1
|x|
2
_
as |x| (5.10)
for a function A dened on the product S
2
S
2
. This function is called the
scattering amplitude.
The inverse obstacle problem: to determine the obstacle K from knowl-
edge of the scattering amplitude (or from a partial knowledge).
Application: radar imaging.
Another kind of obstacle without sharp boundary surface is a non-homogeneity
in the medium, i.e. a variable wave velocity and hence variable refraction coe-
cient n = n(x) . Suppose that the function n is smooth and is equal to a constant
n
0
in X\K. Then again, for arbitrary unit vector there exists a eld u = u
i
+u
s
satisfying the Helmholtz equation
_
+
2
n
2
(x)
_
u = 0
where k = n
0
and the scattered eld u
s
is of the form (9)-(10).
The inverse acoustic problem: to determine the function n from knowl-
edge of a.
Application: ultrasound tomography.
Uniqueness theorems are proved. There is no analytic solution. For the inverse
obstacle problem there are various reconstruction algorithms.
5.5 Interferation and diraction
Take Helmholtz-Kirchhof formula
u(x) =
_
K
_
f (y)
exp (k |x y|)
|x y|
+g (y)
exp (k |x y|)
|x y|
_
dS
y
= v+w, f = u, g =
u
6
Suppose that K is the half-plane {y
1
0, y
2
R} and study the behaviour of
the wave eld near the light-shadow plane L
.
= {x
1
= 0} . Consider the second
integral
w(x) =
_
_
0
g (y)
exp (k |x y|)
|x y|
dy
1
dy
2
We observe the amplitude |w|
2
of the wave eld w on the screen S = {x
3
= r} .
Suppose that g is a C
1
-function decreasing at innity. We can write g (y) =
g
0
(y) +y
1
h(y) for a continuous functions g
0
and h such that g
0
does not depend
on y
1
for o y
1
1
w(x) =
_ _
0
g
0
exp (k |x y|)
|x y|
dy
1
dy
2
+W (x)
where W has a smoother singularity at L. We have |x y| = 1+1/2
_
(x
1
y
1
)
2
+ (x
2
y
2
)
2
_
+
O
_
(x y)
4
_
. The above integral can be aproximated by the product
_
exp
_
k/2 (y
2
x
2
)
2
_
g (0, y
2
) dy
2
_
0
exp
_
k/2 (y
1
x
1
)
2
_
dy
1
where
_
0
exp
_
k/2 (y
1
x
1
)
2
_
dy
1
is called the Fresnel integral. The rst factor
is a smooth function of x
2
according to the stationary phase formula:
_
exp
_
k/2 (y
2
x
2
)
2
_
g
0
(y
2
) dy
2
=
_
2
k
g
0
(x
2
) +O
_
1
k
k
_
A similar representation is valid for the rst term v, hence the amplitude |u| =
|v +w| oscilates near the light-shadow border: the Fresnel diraction:
Huygens-Fresnel Principle: the wave eld generated by a hole in a screen
can be obtained by superposition of elementary elds with the source points in
the hole
References
[1] R.Courant, D.Hilbert: Methods of mathematical physics
[2] M.E.Taylor: Partial dierential equations II
7
Chapter 6
Geometry of waves
6.1 Wave fronts
The wave equation in a non-homogeneous non-isotropical time independent medium in
V = X R is
a (x, D
x
, D
t
) u = f, where (6.1)
a (x, D
x
, D
t
)
.
=
2
t
ij
g
ij
(x)
i
j
+
b
i
(x)
i
+ c (x)
where
i
= /x
i
, i = 1, 2, 3, the functions g
ij
, b
i
, c are smooth in a domain D X and
full the condition
g
ij
(x)
i
j
v
2
0
(x) ||
2
, ||
2 .
=
2
i
for a positive function v
0
. The medium is called isotropic if this is an equation for a
function v
0
which is called the local velocity of the wave. The principal symbol of the
equation is
2
(x; , ) =
2
+ g
ij
(x)
i
j
A wave front is a hypersurface W D that is equal to the singularity set of a solution
u of (1) for some f C
2
(x; ,
t
) = g
ij
(x)
i
j
(
t
)
2
= 0
This is called the eikonal equation, any function satisfying this equation such that
t
= 0
is called an eikonal function. In particular, if (x, t) = t +(x) , the eikonal equation is
g
ij
(x)
i
j
= 1. For isotropical case || = n(x) .
1
6.2 Hamilton-Jacobi theory
Consider the rst order equation in space-time V = X R of dimension n
h(x, t; ) = 0 (6.2)
where h(x, t; ) is a function that is homogeneous in = (, ) . Write the initial condition
as follows (x, t
0
) =
0
(x) .
To solve this equation we consider the system of equations in the phase space V V
:
h(x, t; ) | = 0, | = 0 (6.3)
where = dx+dt is the contact 1-form, is unknown n-dimensional conical submani-
fold in the phase space. (A submanifold in V V
is called conical, if it is invariant under
the mapping (x, ) (x, ) for any > 0. The unknown manifold is Lagrangian, since
of the second equation. Suppose that the form dt does not vanishes in and the following
initial condition is satised:
|
t=t
0
=
0
, where
0
X V
0
of the point (x
0
, t
0
,
0
/
0
)
1
. Therefore we have
j
=
j
(x) in
0
for some smooth
functions
j
, j = 1, ..., n. We have dx + dt = 0 in
1
, hence d dx = 0. It follows that
there exists a function = (x) in a neighborhood of the point (x
0
, t
0
) p (
0
) such
that d =
j
dx
j
|
0
:
(x) = t
0
+
_
j
(x) dx
j
where is an arbitrary 1-chain that joins x
0
with x (i.e. = [x] [x
0
]). Therefore
= d ((x) + t) . This form vanishes in
0
, hence t + (x) = const in and in the
image of in V. We have t
0
+ (x
0
) = 0, hence t + (x) = 0 in
0
. Then the rst
equation (3) implies (2). Now we solve (3)
h(x, t; ) | = 0, | = 0 (6.4)
Reminder: The contraction of a 2-form by means of a eld w is the 1-form w such
that w (v) = (w, v) for arbitrary v.
The Hamiltonian tangent eld v is uniquely dened by the equation
v d = dh
Since d = d dx, this is equivalent to
v = (h
, h
, h
x
, h
t
)
2
which is the standard form of the Hamiltonian eld. We have v (h) = 0. Assume that
h
+ h
) d (dh) = d (h
+ h
) = mdh = 0
We have h
+h
,
dt
ds
= h
;
d
ds
= h
x
,
d
ds
= h
t
(6.5)
where h
dx
ds
+
dt
ds
= h
(x; , ) + h
(x; , ) = h
(x; ) = mh(x; ) = 0
If the Hamiltonian function does not depend on time, we have d/ds = 0.
Construction of wave fronts. Take a wave front W
0
at the time t = t
0
and
consider all the trajectories of the Hamiltonian system that start at a point (x, t
0
; , 1) ,
where x W
0
, is a covector that vanishes in T
x
(W
0
) and fulls the eikonal equation,
i.e. h(x, t
0
; , 1) = 0. The union of these trajectories is just the front W in the domain
{t > 0} .
The condition p : X is of maximal rank may be violated somewhere. Then the
wave front get singularity and the corresponding solution has a caustic.
Proposition 3 2 If a characteristic surface W is tangent to another characteristic sur-
face W
that contains p.
Proof. Let = (, 1) the normal covector at p to both surfaces. According to the
above construction the front W is the projection of a conic Lagrange manifold , which
is a union of trajectories of (5) and W is the union of rays. The point (p, ) belongs to
, since the form dx + dt vanishes in T (W) . Let be the ray through p and be the
corresponding bicharacteristic strip through (p, ) . It is contained in since any solution
of (5) is dened uniquely by initial data. Therefore W and similarly W
.
6.3 Geometry of rays
If h does not depend on x and t, the trajectories are straight lines. One more case when
the rays can be explicitly written is the following
Proposition 4 If the velocity v is a linear function in X, the rays in the half-space
{v > 0} are circles with centers in the plane {v = 0} .
Problem. To check this fact.
3
6.4 Legendre transformation and geometric duality
Denition. Let f : X R be a continuous function; the function g dened in X
by
g ()
.
= sup
x
x f (x)
is called Legendre transformation of f. If f is convex, g is dened in a convex subset
of the dual space and is also convex. If g is dened everywhere in X
, the Legendre
transformation of g coincides with f, provided f is convex. This means that the graph
of f is the envelope of hyperplanes t = x g () , X
.
If f C
1
(X) an arbitrary function, the Legendre transformation is dened as follows
g () = x f (x) as = f (x)
If f C
2
(X) and det
2
f = 0 the equation f (x) = can be solved, at least, locally
and the Legendre transform is dened as a multivalued function.
Example. For a non-singular quadratic form q (x) = q
ij
x
i
x
j
/2 the Legendre trans-
form is again a quadratic form, namely, q () = q
ij
j
/2, where {q
ij
} is the inverse
matrix to {q
ij
} . Indeed, the system
i
=
i
q (x) = q
ij
x
j
is solved by x
i
= q
ij
j
. Then the
Legendre transform equals
i
q
ij
j
q
ij
q
ik
q
jl
l
/2 =
i
q
ij
j
/2 = q ()
Denition. Let K be a compact set in X. The function
p
K
() = max
K
x
is called Minkowski functional of K. If K is convex and simmetric with respect to the
origin, the functional p
K
is a norm in X
K
() 1} is equal to the norm p
K
in X generated by K.
Problem. Show that the Legendre transform of the function (p
K
)
2
/2 is equal to
(p
K
)
2
/2, provided K is convex.
Denition. Let Y be a smooth conic hypersurface in X (i.e. Y is smooth in X\ {0}).
The set Y
(r))dr
is called the optical length of the curve (or the action). It is equal to the time of a
motion along with the velocity
_
q (x(r) , x
(r)) (v = n
1
in the isotropical case).
Proposition 5 3 Each ray of the system (5) for the Hamiltonian function h =
2
/2 is
an extremal of the optical length integral T () .
Proof. We compare the Euler-Lagrange equation
d
dr
F
x
F
x
= 0 (6.6)
for F =
_
q (x, x
) with the system (5). Suppose for simplicity that the medium is
isotropic, i.e.
_
q (x, x
) = n(x) |x
| , h(x; , ) =
_
v
2
(x) ||
2
2
_
/2. Set
=
F
x
= n(x)
x
|x
|
,
d
ds
=
1
n|x
|
d
dr
The Euler-Lagrange equation turns to
d
ds
=
1
n|x
|
d
dr
F
x
=
1
n|x
|
F
dx
=
n
n|x
|
= v
2
||
2
/2 = h
x
since || = n, whereas
dx
ds
=
x
n|x
|
= v
2
= h
2
(x; ) = g
ij
(x)
i
j
2
, = (, )
is the principal symbol of the equation (1). Fix x and consider the cone K
x
.
= {
2
(x; ) = 0}
in V
h(x; y, y
0
) =
_
g
ij
(x) y
i
y
j
y
02
_
/2
5
is the Legendre transform of h
.
=
2
/2 and (y, y
0
) stands for a tangent vector to V at
(x, t) .
The major Huygens principle. Let W
0
be a smooth wave front at a moment
t = t
0
. For a small time interval t and an arbitrary point x W
0
take the ellipsoid
S
x
.
=
_
h(x; x, t) = 0
_
(6.7)
Let
W be the envelope of these ellipsoids. The claim: the wave front W
t
at the moment
t = t
0
+ t coincides with a component of
W up to O(t
2
) .
This means, in fact, that the distance between the hypersurfaces is O(t
2
) in the
standard C
1
-metric.
Proof. An arbitrary point x W
0
is the end of a ray given by an equation
x = x(t) , 0 t t
0
. According to (5), the extension of this ray for the time interval
[t
0
, t
0
+ t] is approximated by the line interval [x, x] , where x = x + t x
, x
=
h
h(x, t x
, t) =
h(x, x
, 1) , since
h is a homogeneous quadratic function. By the involutivity of the
Legendre transform,
h is the Legendre transform of h, i.e.
h(x, x
, 1) =
x
+ h
_
x;
,
_
where the point
_
,
_
satises h
_
x;
,
_
= x
, h
h(x, x
, 1) = x
+ 1 h(x; , 1) = h
(x; , 1) + h
is less that t + t since of the Ferm at principle. Therefore this point is behind the front
W
t
. This completes the proof.
6.7 Geometrical optics
This is the ray method (Debays method) and similar methods for construction of high
frequency approximations to solutions of the wave equation:
au
= O
_
q
_
where a is a wave operator (1) or a similar operator. One looks for an approximate
solution of the form (WKB-form)
u
2
+ g
ij
j
+ b
j
j
+ c
_
U (x, ) = O
_
k
_
The phase function satises the eikonal equation 1 g
ij
j
= 0 and the amplitude
functions a
0
, a
1
, ..., a
k
full the recurrent dierential equations, called transport equations
2g
ij
j
a
0
+
_
i
g
ij
j
() + b
j
_
a
0
= 0 :Ta
0
= 0 (6.9)
2g
ij
j
a
1
+
_
i
g
ij
j
+ b
j
_
a
1
=
_
i
g
ij
j
+ b
j
j
+ c
_
a
0
...
Ta
k
= L
k
(a
0
, ..., a
k1
)
where the operator T = 2g
ij
j
+ (
i
g
ij
j
() + b
j
j
) acts along geodesic curves of
the metric g. The principal term of (8) is called the approximation of geometrical optics.
A caustic is an obstruction of the ray method.
6.8 Caustics
Take the manifold in the phase space that is solution of the system (3)
h(x, t; ) | = 0, | = 0
If dim = n = dimV, it is called Lagrange manifold. Consider the projection p : V ;
the image W = p () is called wave front. A point (x, t) W is regular, if (x, t) =
p () , and rank of p in is equal n 1 for any . The set of singular points is
closed; its projection to X called the caustic of .
6.9 Geometrical conservation law
We have found the conservation law for the global energy of a eld u satisfying the
selfadjoint wave equation (MP2) by means of
t
_
2
u
t
2
, u
_
=
t
_
x
i
_
v
2
u
x
i
_
, u
_
This identity can be written in the form
div I
x,t
= 0, where I
x,t
.
=
_
v
2
(
x
uu
t
u
t
u) , u
t
u
t
_
The space-time eld I
x,t
is interpreted as the energy current. For a time-harmonic solution
u(x, t) = exp (t) U (x, ) the last component drops out and u
t
= u. Therefore the
energy current is represented by the eld I
x
= v
2
(
x
uu
t
u
t
u) . For the arbitrary
selfadjoint wave operator
_
2
t
i
g
ij
j
c
_
u = 0
7
the energy current is
I
i
=
2
1
g
ij
_
j
UU U
j
U
_
, i = 1, 2, 3
Substitute the WKB-development for (8) and take in account that the phase and ampli-
tude functions are real:
I
i
=
2
g
ij
j
a
0
+ O()
The vector g
ij
j
= h
i
(x, ) = h
i
(x, ) = dx
i
/ds is equal to the tangent to the ray
through a point x X. It follows
Corollary 8 The energy ows along the rays in the approximation of geometrical optics.
This fact can be explained in a dierent way. Consider the transport equation for the
main term of the amplitude
Ta
0
.
= 2g
ij
j
a
0
+
i
g
ij
j
() a
0
= 0
and write it in the form
2da
0
/ds +
i
_
v
i
_
a
0
= 0 (6.10)
where
i
v
i
= div v (s) , and v
i
.
= h
i
= g
ij
j
is the X-component of Hamiltonian eld
that generates the geodesic ow (5). We have div v (s) = (V (s))
1
dV (s) /dswhere V (s)
is the image of the volume element dx in X. Indeed, we have
L
v
(dx) = d (v dx) = div (v) dx
Therefore (10) is equavalent to
d
ds
_
a
0
dx
_
= 0
i.e. the halfdensity a
0
a
0
dx
2
= |a
0
|
2
dx of the wave eld.
Corollary 9 The energy density is preserved by the geodesic ow.
Another conclusion is: a solution of the Helmholtz equation can be considered a half-
density, whose square is the energy density. Also the halfdensity a
0
dx dt is preserved
by this ow since dt is constant, since v
t
= h
t
= 0. Therefore a solution of the wave
equation is a halfdensity in space-time.
8
Chapter 7
The method of Fourier integrals
7.1 Elements of simplectic geometry
Cotangent bundle. Let M be a manifold. Consider the set T
(M) =
M
T
x
(M)
together with the mapping p : T
x
(M) to the point
x. It maps an arbitrary element T
x
(M) to the point x. The pair T
(M), p) is
called cotangent bundle of the manifold M. The bundle possesses a smooth atlas: for an
arbitrary chart (U, ) in M one takes the set T
(M).
Each element T
x
(M) can be written in the form =
m
1
j
dx
j
, where the coecients
i
are uniquely dened. The mapping
p : p
1
(U) R
m
R
m
, () = (
1
, ...,
m
) (7.1)
is a chart in T
= ,
where the transition mapping is of the form ((
1
, ...,
n
), x) = (
1
, ...,
n
), x). Here
i
are coecients of cotangent vectors in the second chart: =
i
dx
i
. They are related
to the coecients in the rst charts:
j
() =
i
x
j
i
()
Here J = {x
i
/x
j
} is again the Jacobi matrix of the transition mapping . Consequently
the relation between the coecients is linear and smooth with respect to the coordinates
in U (as well as in U
and
T
(M) M is a mapping
of smooth manifolds. Each bre p
1
(x) = T
x
(M) is a vector space (hence T
(M) is a
vector bundle).
Remark. The union of sets T
x
(M), x M has a structure of vector bundle too. It is
called tangent bundle.
Canonical forms. The 1-form
U
=
i
dx
i
is dened for each chart (1). For another
chart in p
1
(U
) the forms
U
and
U
coincide in the intersection p
1
(U) p
1
(U
). This
follows from (2). Therefore there is well-dened a 1-form in T
(M).
The form = d is called canonical 2-form in the simplectic manifold T
(M). It is
closed: d = 0. In local coordinates =
m
1
d
i
dx
i
.
1
Denition. Let M be a manifold of dimension m. A submanifold T
(M) is called
Lagrange manifold, if it satises the conditions dim = m and |N = 0.
Proposition 1 1 Let be a Lagrange manifold in T
i
=
f
x
i
, i = 1, ..., m (7.2)
coincides with in a neighborhood of .
Proof. Let U be a simply connected neighborhood U of y such that the projection
p is a dieomorphism p
U
: (U) U, where we denote (U) =
1
(U). Take a
point (U) and join the point x = p() with the point y by a curve
x
U. We lift
this curve by means of the mapping (p
U
)
1
and get a curve (U), which join the
point with . The integral f() =
_
0
(M) the open subset in T
0
(M) as follows t :
(x, ) (x, t). A trajectory of the group
+
is called ray. A subset K T
0
(M) is called
conic, if K is invariant with respect to the group, i.e. is a union of rays. We note that no
conic Lagrange manifold can satisfy the conditions of Proposition 1. We generalize this
proposition in the next section.
Proposition 2 2 A conic submanifold of dimension dimM is a Lagrange manifold if
and only if the canonic 1-form vanishes in .
Proof. The part if is obvious: |K = d|K = d(|K) = 0. We need to check that
the equation | = 0 implies | = 0. Consider the eld e =
i
/
i
(Euler eld)
in the cotangent bundle. It satises the equation e = . The Euler eld is tangent
to rays and hence to any conic submanifold. Therefore for any eld v in T
(M) that is
tangent to we have
(v) = v = v (e ) = (e, v) = 0
Example. Let P be a submanifold of manifold M. Consider the set N
P
(M) T
(M) of
points (x, ), x P such that the form
i
dx
i
vanishes in T
x
(P). It is called conormal
bundle to P. This is obviously a conic Lagrange manifold.
2
7.2 Generating functions
We state a generalization of Proposition 1 for the case of critical point of the projection
p : M. First we state
Proposition 3 3 Let be Lagrange manifold in T
() T
y
(M) Suppose that the forms p
(dx
1
), ..., p
(dx
r
)
are independent in T
of the point .
Proof. The statement follows from the implicit function theorem, if we show that
the point is not critical for the mapping . Suppose the opposite. Then there exists a
tangent vector t T
r+1
a
i
x
i
+
r
1
b
j
j
Show that the coecients a
i
are equal zero. In virtue of the assumption for each i =
r + 1, ..., n the restriction of the form dx
i
to the space T
() = 0, where
.
= dx
i
r
1
c
j
dx
j
. Therefore we have
0 = (t) = a
i
.
The form t =
b
j
dx
j
is vanishes in T
1
=
f
x
1
, ...,
r
=
f
x
r
(7.4)
x
r+1
=
f
r+1
, ..., x
m
=
f
m
in a neighborhood of the point
0
.
Proof. We have = d
, where
r
1
i
dx
i
m
r+1
x
j
d
j
. Choose a simply
connected open set W R
r
R
mr
such that the projection :
1
(W) W is a
dieomorphism and set
f(x
) =
_
, x
= (x
1
, ..., x
r
),
= (
r+1
, ...,
n
)
where is an arbitrary curve in
1
(W) that connects
0
and
.
=
1
(x
). The
integral does not depend on the curve in virtue of the equation d
| = | = 0. We
have df =
0
= (x
0
,
0
). Dene a generating function f by means of the integral as above taken
over a curve from the point (x
0
, 0) to . This is a generating function too. Check that
it is homogeneous. Really for any = (x, ) we can take the curve = g r, where g
joins the points x
0
and x and r is the ray {(x, t), 0 t 1}. Therefore
f(()) =
_
=
_
r
=
m
r+1
x
j
j
,
where x
r+1
, ..., x
m
are functions of x
1
, ..., x
r
.
7.3 Fourier integrals
Fourier integral in an open set X R
n
is a functional of the form
I(, a){} =
_
X
_
= S
D
i
x
D
j
a(x, )
C
ij
(x)(|| + 1)
|j|
, (7.7)
with a continuous function C
ij
that does not depend on . We call the number
.
= +N/2
order of the Fourier integral I(, a).
4
Example. An arbitrary smooth homogeneous amplitude a of degree belongs to S
1
.
Denition. We say that an amplitude a is asymptotical homogeneous of order , if it
has for any q the following development:
a = a
+a
1
+... +a
q
+r
q
,
where each term a
|i|q
_
X
C(x)
D
i
(x)
dx
for a continuous function C. In particular, (6) implies that the distribution I(, a) is of
singular order 0.
Theorem 6 Let be arbitrary smooth real homogeneous of degree 1 function in X
without critical points and a S
j
, j = 1, ..., N
are homogeneous operators of degree 1, if the functions b(x, ) and c(x, ) are homo-
geneous of degree 1 and 0 respectively. The condition (7) implies that for arbitrary
homogeneous operator A of degree 1 and amplitude a S
we have Aa S
.
Pick out a function D(R
N
) such that () = 1 for || 1. Write (5) in the form
of sum of two integrals with the extra factors and 1 . The rst one is a proper
integral which converges to [I
0
dx]() as 0, where
I
0
(x) =
_
exp(2(x, ))()a(x, )d
is a continuous function. Set
= 2 +||.
5
Lemma 7 There exists a smooth family of tangent elds v
, 0 of degree 1 in X
such that v
) = .
We postpone a proof of this Lemma. Write the second integral as follows
_ _
exp(
)(1 )adxd =
_ _
v
(exp(
))(1 )adxd
Integrating by parts the right side, we get the integral
_ _
exp()v
((1 )a)dxd,
where v
((1 )a) = [v
() div(v
(a) +av
()),
whence
_ _
exp(
)(1 )adxd =
_ _
exp(
)
_
a
0
+
j
a
j
x
j
_
dxd, (7.9)
and
a
0
= [v
() div(v
)(1 )]a +v
(a), a
j
= av
(x
j
), j = 1, ..., n
The amplitude a
j
, j = 1, ..., n belong to the class S
1
(x
j
) is smooth and homogeneous of
degree 1. The same is true for the function [v
() div(v
) is
homogeneous of degree 1 and v
(a) belongs to
the space S
and satisfy the corresponding inequality (7) with some constants that do
not depend on . Taking in account the inequality 1, we conclude that the functions
a
0
, ..., a
n
satisfy (7) with some uniform constants and with the exponent instead of
. If +N < , the integrals
_
exp(
)|a
j
|d, j = 0, 1, ..., n
converges uniformly with respect to and we can pass on to the limit in (9). Thus we
get the inequality
| lim
0
_
_
X
exp(
)a(x, )(x)dxd C
_
_
||dx +
n
1
_
x
j
dx
_
where the constant C does not depend on . It follows that I(, a) is a distribution of
order 1.
If the opposite inequality + N holds, we apply the same method to each term of
(9) and get
I(
, a)() =
_
exp(
ij
_
a
00
+
a
0j
x
j
+
a
ij
x
i
x
j
_
dxd,
6
where the amplitudes a
ij
belong to S
2
b
j
x
j
+
c
i
i
,
where
b
j
=
x
j
, c
i
=
||
2
i
, =
x
j
2
+||
2
2
The dominator does not vanish. We have v(
= (1 +v(||))
1
v.
Lemma 8 Let u be an arbitrary homogeneous tangent eld in X of degree 1 and
a smooth dierential form of the highest degree in X that vanishes in the complement
of K for a compact set K X such that the forms and L
u
are integrable. We
have
_
L
u
() = 0 (7.10)
Proof. Suppose that the form has compact support and state the equation
_
t
() =
_
(7.11)
for small t, where
t
means the ow generated by the eld u. The integral of a form of
the highest degree is invariant with respect to any isomorphism of the manifold. Take the
t-derivative of (11) and get (10). For the given form we consider the product
k
= h
k
where h
k
() = h(k
1
). The integral
_
k
converges to
_
as k . We have
0 =
_
L
u
(
k
) =
_
u(h
k
) +
_
h
k
L
u
()
We have
_
h
k
L
u
()
_
L
u
() as k . At the other hand u(h
k
) =
c
j
h
k
/d
j
=
O(k
1
) uniformly in X , since the functions c
j
= u(
j
), i = 1, ..., N are homogeneous
of degree 0. Therefore
_
u(h
k
) 0.
Example. Let X be an open set in R
n
, x
1
, ..., x
n
are coordinate functions. Take =
fgdx d in (10) and get
_
u(f)gdxd =
_
fu
(g)dxd, (7.12)
where the sum
u
= u div u, div u
L
u
(dx d)
dx d
=
b
j
x
j
+
c
i
i
7
is the conjugated dierential operator to the eld u. We check the last equation by means
of (4.4.8):
L
u
(fgdx d) = u(f)gdx d +fu(g)dx d +fgL
u
(dx d)
L
u
(dx d) = d(u (dx d)) =
j
(1)
j1
d(b
j
dx
1
... ... dx
n
d)
+
i
(1)
n+i1
d(c
i
dx d
1
...... d
N
) = div(u) dx d
Remark. We can use instead of E
= exp(||
2
) in (8) . We get the same limit.
Non-degenerate phase. Let be a phase function in X . Consider the set
C() = {(x, ) : d
= 0,
1
= ... =
N
= 0}
where
j
.
= /
j
. This is the critical set of the projection { = 0} X.
Denition. The phase function is called non-degenerate, if it has no critical points
and the dierential forms
d
_
1
_
, ..., d
_
N
_
are linearly independent in each point of the set C(). Suppose that is a non-degenerate
phase. The critical set C() is a conic subset of X of dimension n + N N = n =
dimX. This follows from the Implicit function theorem. Recall that T
: C() T
= 0. This mapping is
homogeneous of degree 1, since d
x
(x, t) = td
x
(x, ) for t > 0.
Proposition 9 The dierential D
: T(C()) T(T
is injec-
tive in each point of C().
Proof. The injectivity of D in a point (x, ) C() is equivalent to the following
implication:
v T
(x,)
(C()), D
(v) = 0 =v = 0
Write v = t +, t T
x
(X), T
(v) =
_
t; v(
x
1
), ..., v(
x
n
)
_
T
(T
(X)) (7.13)
We denote here = (x, d
x
(x, )) and use the natural isomorphism
T
(T
(X))
= T
x
(X) R
n
From (12) we conclude that t = 0 and (
x
j
) = 0, j = 1, ..., n At the other hand the
vector = v is tangent to C(), which means
(
i
) = 0, i = 1, ..., N
8
Extend the vector to the constant vector eld in . It commutes with the coordinate
derivatives in X , consequently the last equations are equivalent to the following
d
1
, ..., d
N
. This relation is
in fact trivial, since the phase is non-degenerate.
Denote by () the image of the mapping D
to X
0
0
is a dieomorphism
to its image
0
. We can take for
0
a conic neighborhood since the mapping D
is
homogeneous. The image
0
is a conic submanifold of dimension n = dimX; it is closed
in a conic neighborhood of the point
0
=
(x
0
,
0
). The variety () is a union of
pieces
0
, hence it is a conic set too. If a neighborhood X
1
1
overlaps with X
0
0
,
then its image
1
is a continuation of the manifold
0
. Taking a chain of continuations
0
,
1
, ... we can reach a self-intersection point, if the mapping D
is not an injection.
In this case the set L() may have singular points and we call it variety.
Proposition 10 The set () is closed and locally equal a nite union of conic Lagrange
manifolds.
Proof. Show that the canonical 1-form vanishes in any vector w T
(x,)
(T
(X)),
which belongs to the image of a tangent space T
(x,)
(C()). We have = d
x
(x, ) and
w = D
(v) for a tangent vector v to C() at the point (x, ). Therefore v(f) = 0 for
arbitrary function f that vanishes in C(). Let t be the projection of w to X; it is equal
the projection of v. We calculate
(w) = dx(t) = d
x
(t) = t() = v(),
where the righthand side is taken at the point (x, ) C(). It is equal zero, because of
the function vanishes in C(). The last fact follows form the Euler identity =
i
.
It follows that any piece
0
of the set () is a Lagrange manifold.
Take an arbitrary point (), a neighborhood U of the point x = p() such that its
closure K is compact and check the set
K
.
= ()p
1
(K) is closed. For this we take the
unit sphere S() in the ancillary space and consider the mapping
: C() (KS())
L
K
). It is continuous and the rst topological space is compact. Therefore the image
is a closed subset of (). The conic set L
k
() is generated by this subset and hence is
also closed.
Show that
K
is covered by a nite number of Lagrange manifolds. The set K can
be covered by a nite number of conic neighborhoods X
q
q
, q = 1, ..., Q as above.
The restriction of the mapping
(X
q
q
), which implies our assertion.
Proposition 11 Let be a conic Lagrange manifold. For any point there exists
a non-degenerate phase function such that () .
Proof. Take the generating function f =
m
k+1
f
j
j
at constructed in Proposition
4.8.2 and consider = (
k+1
, ...,
m
) as ancillary variables. Here f
j
= f
j
(x
, ), j =
k + 1, ..., n are smooth functions in W such that the equations
x
j
= f
j
(x
, ), j = k + 1, ..., m (7.14)
9
are satised in . We set
(x, )
.
=
x
j
j
f(x
, ) =
m
k+1
(x
j
f
j
)
j
and have /
j
= x
j
f/
j
= x
j
f
j
(x
(x
, ), ) = |
7.4 Lagrange distributions
Denition. Let X be an open set in R
n
and be a closed conic Lagrange submanifold
in T
I(
j
, a
j
) +v, v C
,
where for each j the phase
j
is non-degenerate in X
j
and
(
j
) , a
j
S
(X
j
)
Denition. Suppose that all amplitudes are asymptotical homogeneous. We shall say
that the Lagrange distribution u is of order , if u admits such a representation where
all the Fourier integrals I(
j
, a
j
) are of the order .
Example 5.2. Let Y be a closed submanifold of X given by the equations f
1
(x) =
... = f
m
(x) = 0 such that the forms df
1
, ..., df
m
are independent in each point of Y .
Consider the functional
Y
() =
_
Y
df
1
... df
m
on the space D(X) of test densities. The quotient is a density in Y such that df
1
...
df
m
= , hence the integral is well-dened. It is called the delta-function in Y .
Show that the delta-function is a Fourier integral with N = m if X is an open set in
R
n
. Take the phase function (x, ) =
m
1
j
f
j
(x) and the amplitude a = 1. In the case
n = 1 we have for any test density = dx
I() =
_ _
exp(2f(x))d(x)dx =
_
exp(2y)d
f
dy,
if we take y = f(x) as an independent variable. The -integral is equal the delta-function,
hence I{} = /df|f = 0, where /df is a smooth function. In the case m > 1 we use
this formula m times and get
I{} =
_ _
exp(2(x, ))d =
_
Y
df
1
... df
m
=
Y
()
where
Y
is the delta-function in the manifold Y . This is a -distribution of order
(dimX dimY )/2 for = N
Y
.
10
Properties. For a conic Lagrange manifold we denote D
().
If u is of order , then Pu is of order +m, where m is the order of a.
III. Restriction to a submanifold. Let Y be a closed submanifold in X such that
N
(Y ) = . Denote
Y
= {(y, ) : y Y, = |T
y
(Y ), (y, ) }
This is a conic Lagrange submanifold in T
(Y ).
Proposition 12 Any -distribution u has a restriction u
Y
that is a
Y
-distribution. If
u is of order , the distribution u
Y
is of order too.
IV. Product. If
-distribution u
the product uu
is well-
dened as a distribution in X.
7.5 Hyperbolic Cauchy problem revisited
Consider a hyperbolic dierential equation of order m in a space-time X = X
0
R, where
X
0
is an open set in R
n
a(x, t; D
x
, D
t
)u = w (7.15)
with smooth coecients in X; x = (x
1
, ..., x
n
) are spacial coordinates, t is the time
variable. We denote by , the corresponding coordinates for cotangent spacial and time
vectors respectively. The principal symbol
m
=
m
(x, t; , ) of (14) is a polynomial in
variables , . We suppose that it has order m with respect to , which means that any
hypersurface t = const is non-characteristic for P. Consider the Cauchy problem in the
domain t > 0 with the initial data
u(x, 0) = v
0
(x),
u(x, 0)
t
= v
1
(x), ...,
m1
u(x, 0)
t
m1
= v
m1
(x), (7.16)
where v
0
, ..., v
m1
are some distributions.
Theorem 13 (Uniqueness) Any strictly hyperbolic Cauchy problem (14),(15) has no
more than one solution.
Fix a point y X
0
; let E
j
y
D
(XR
+
), j = 0, ..., m1 be the solution of the initial
problem with w = 0, v
i
=
i
j
y
. The set of distributions E
0
y
, E
1
y
, ..., E
m1
y
in X
+
X
is called fundamental solution of the Cauchy problem. Then one can solve the Cauchy
problem with w = 0 and arbitrary distributions u
0
, ..., u
m1
by means of integration:
u =
k
_
X
0
E
k
y
v
k
(y)dy
11
This formula is valid, at least, for distributions v
k
with compact support. In the global
case we need an assumption on domain of dependence (see below). The general case is
reduced to the case w = 0 by means of the Duhamels method.
Remark. If the coecients of the operator a do not depend on time, it is sucient to
construct the distribution E
m1
y
only, since we have E
k
y
= q
m1k
(y, D)E
m1
y
, k < m1,
where q
j
is an appropriate dierential operator of order j. Then the distribution E
y
=
E
m1
y
is called the fundamental solution.
We describe now a more general construction. Therefore we can represent the symbol
as the product of binomials:
m
(x, t; , ) = q
0
(x, t)
m
1
[
j
(x, t; )],
where
1
, ...,
m
are homogeneous functions of variables of degree 1 and q
0
= 0. Let
0
T
(X
0
) be an arbitrary Lagrange manifold. For any number j = 1, ..., m we consider
the Hamiltonian function h
j
(x, t; , ) =
j
(x, t; ) in T
(X R
+
). We lift
0
to
the bundle T
,
d
dt
=
h
j
x
d
dt
=
h
j
t
(7.17)
with initial data from W
j
. Denote by
j
the union of trajectories of this ow. This is a
Lagrange manifold
j
in T
, ) for
0
, where x
= (x
1
, ..., x
r
) and = (
r+1
, ...,
n
),
N = nr. Let (x
0
,
0
) be the coordinates of and x
0
.
= p() X
0
. Take a phase function
0
=
0
(x, ) in a conic neighborhood of (x
0
,
0
) that generates
0
in a neighborhood of .
We can write the initial data v
0
, ..., v
m1
as Fourier integrals with the phase function
0
and some asymptotical homogeneous amplitudes b
0
, ..., b
m1
in a neighborhood of (x
0
,
0
),
where b
k
is of order N/2 + k for k = 0, ..., m 1. The functions (x
, t; ) form a
local coordinate system in
j
for any j and we can choose a generating phase function in
the form
j
such that
j
(x, 0; ) =
0
(x, ). Set u
j,
.
= I(
j
, a
j
), where a
j
are unknown
homogeneous amplitudes of degree N/2 and substitute it in the equation. We get a
-distribution w = au
j,
with the symbol
(w) =
j
(L +s) (u
j,
),
12
where L = L
p
m
is the Lie derivative. The term of degree + m vanishes according to
Proposition 5.6.1 since the symbol
m
=
h
k
vanishes in
j
. The next term is calculated
by means of Theorem 6.1.1. where s is the subprincipal symbol of P. The degree of this
term is equal + m 1. We choose the amplitudes a
j
in such a way that the symbol
of w vanishes. For this we solve rst the equations (L +s) (u
j
) = 0. According to
(5.5.1) we have (u) = a
j
j
, where
j
is a non-vanishing halfdensity depending only on
the phase function
j
and a
j
be the principal homogeneous term of A
j
of degree N/2.
Dividing the above equation by this halfdensity we get an equation
L(a
j
) +g
j
a
j
= 0 (7.18)
where g
j
is a known function. This is an ordinary equation along the trajectories of the
eld (16). It has a unique solution for an arbitrary initial data a
j
(x, 0; ). We specify
these data to satisfy the initial condition (15) for the Cauchy problem. This gives the
equations
(2)
k
j
(
j
)
k
a
j,
(x, 0; ) = g
(x, )b
k
(x, ), k = 0, ..., m1, (7.19)
where we denote
(X
0
) is compact). In the k-the equation both sides are homo-
geneous of the same degree N/2 + k. To solve this system we consider the matrix
W = {(
j
)
k
}, where
= d/dt. We have
det W =
j<k
(
k
),
hence the matrix W is invertible, if
j
=
k
for j < k. We have
j
=
j
(x, t; d
x
j
),
since the function h
j
vanishes in
j
. Therefore d
x
j
(x, 0; ) = d
x
0
(x; ) = 0 in C(
0
).
The functions
j
(x, 0; d
x
0
), j = 1, ..., m are dierent, because of the operator is strictly
hyperbolic and d
x
0
= 0. Therefore the matrix is invertible and the system (18) has
a solution a
1,
(x, ), ..., a
m,
(x, ), whose components are smooth and homogeneous of
degree N/2 and compactly supported in V . Then we solve the transport equations
(17) with the initial condition a
j,
(x, ) for the j-th equation. The solution a
j,
(x, t; )
exists and is uniformly bounded in a conic neighborhood of in
j
. The Fourier integral
u
j,
.
= I(
j
, a
j,
), j = 1, ..., m satises the equation in the rst and second highest orders,
i.e. the amplitude of Pu
j,
is of of order N/2 + m 2. Set u
u
j,
for an
appropriate partition of unity {g
} in a neighborhood of
0
. This distribution satises
the equation au
= w
1
, where w
1
is a -distribution of order + m 2 and initial
conditions v
k
k
t
u
|t = v
k
, where v
k
is a
0
-distribution of order + k 1 for
k = 0, ..., m1.
For the next approximation we look for a new homogeneous amplitudes a
j,
of degree
N/2 1 and take u
j,
= I(
j
, a
j,
). Calculating the symbol, we nd
(a(u
j,
+u
j,
)) = (L +s) (u
j,
) +q
1
,
13
where q
1
is a asymptotically homogeneous halfdensity of order + m 2 that only
depends on u
j,
. We need to solve the equation
(L +s) (u
j,
) = q
1
in
j
with the initial data taken from the system
(2)
k
j
(
j
)
k
a
j,
(x, 0; ) = g
(x, )b
k
(x, ), k = 0, ..., m1
Here b
k
are some homogeneous amplitudes of degree N/2 + k 1, k = 0, ...m
1. In fact we take for b
k
the principal homogeneous terms of amplitudes of Fourier
integrals representing new initial data v
k
= v
k
k
t
u
j
and set u
1
=
g
,
u
j,
. The sum u
+ u
1
is the second
approximation. It satises the equation P(u
+u
1
) = w
2
, where w
2
is a -distribution
of order +m3 and initial conditions v
k
t
u
1
|t = v
k
, where v
k
is
0
-distribution
of order +k 2.
Iterating these arguments we construct an innite series
u
+u
1
+u
2
+...
of -distributions of orders , 1, .... We can modify the above construction in such
a way that all the amplitudes in the the term U
k
vanish in the ball || k. Then
this series converges to a -distribution u. It satises the conditions: Pu is smooth in
Y and the initial conditions are satised up to smooth functions. Such distribution u is
called parametrix of the problem. To get an exact solution from a parametrix one need
to nd a smooth solution u
(X
f
). This means that the polynomial
m
(x, f(x); , df(x)) is of degree m
with respect to and all his roots are real and dierent. If f decrease suciently fast
at innity the bundle N
(X
f
) has no common points with . Therefore our solution u
has restriction to X
f
and this restriction is a
f
-distribution as well as restrictions of its
conormal derivatives. We take the restriction of the derivatives as new initial conditions
14
in X
f
and solve again the Cauchy problem in a neighborhood Y
f
of X
f
. This solution u
f
agrees with u. They make together a solution of the Cauchy problem in Y
0
Y
f
. Then
we choose a hypersurface X
g
= {t = g(x)} in Y
f
such that g > f and so on. From the
condition of theorem follows that we can regulate this construction in such a way that
the union of all neighborhoods Y
0
, Y
f
, Y
g
, ... coincides with X
0
[0, T).
Take an arbitrary point y X
0
and consider the Lagrange variety T
y
(X
0
). Apply the
construction of Theorem 6.2.1 taking for
0
this manifold. Let
y
be the corresponding
Lagrange manifold over X.
Corollary 16 Suppose that for any compact set K X its domain of dependence is
again a compact set. Then for any y X
0
there exist fundamental system E
0
y
, E
1
y
, ..., E
m1
y
,
where E
k
y
is a
y
-distribution of order (n 1)/2 k.
For each k, 0 k < m we apply Theorem 6.2.1 to the initial data v
k
=
y
, v
j
=
0, j = k. The delta-distribution
y
is a
y
-distribution of order (n 1)/2. Therefore the
solution of the Cauchy problem is a
y
-distribution of order (n 1)/2 k.
Remark. We have WF(E
k
y
)
y
according to Property I of Sec.5.3. Therefore supp E
k
y
is contained in the locus L
y
= p(
y
). The locus is the union of all bicharacteristic curves
starting at y. If the coecients of the symbol
m
are constant, these curves are straight
lines and L
y
is a cone with the vertex at y. In general case the locus L
y
is called ray
conoid.
Another geometrical construction of the conoid can be done in dual terms. Take
coordinates x
1
, ..., x
n
in X
0
that vanish at y and consider the phase function
0
= x
.
=
1
x
1
+ ... +
n
x
n
. It generates the Lagrange manifold T
y
(X
0
). Any phase function
j
has the form
j
(x, t; ) = x +
j
(x, 0; )t + O(t
2
), since h
j
(x, t;
j
,
j
) = 0. For any
= 0 the hypersurface H
j
() = {
j
= 0} is smooth and tangent to the hyperplane
x + (y, 0; )t = 0 at y. Consider the family of varieties H
j
() where ranges in the
unit sphere in T
y
(X
0
) and j runs from 1 to m.
Proposition 17 The conoid L
y
is contained in the envelope of the family {H
j
()}.
Proof. Apply Proposition 5.4.1 to the fundamental distributions:
E
k
y
=
j
_
S()
(
j
(x, ) + 0)
k+1n
a
j
(x, )d,
Here a
j
are smooth functions in U S(), where U is a neighborhood of y. This is true,
if n > k + 1. We see that the kernel (
j
+ 0)
k+1n
is singular only in H
j
(), hence the
integral is smooth in the compliment to the envelope of the family as above. If n k +1
a similar formula holds with the extra factor log |
j
| in the integrand. This implies the
same conclusion.
References
[1] V.P.Palamodov, Lec4.tex
15
Chapter 8
Electromagnetic waves
8.1 Vector analysis
Vector operations: Let X be an oriented Euclidean 3-space X with a frame (e
1
, e
2
, e
3
) .
For vectors U = u
1
e
1
+ u
2
e
2
+ u
3
e
3
, V = .., W = ... X
U V = det
_
_
u
1
u
2
u
3
v
1
v
2
v
3
e
1
e
2
e
3
_
_
= V U
(U V, W) = det
_
_
u
1
u
2
u
3
v
1
v
2
v
3
w
1
w
2
w
3
_
_
U (V W) = (U, V ) W + (U, W) V = (U V ) W
For a smooth vector eld V and a function a
V = rot V = curl V = det
_
_
1
2
3
v
1
v
2
v
3
e
1
e
2
e
3
_
_
= (
2
v
3
3
v
2
)e
1
+ (
3
v
1
1
v
3
)e
2
+ (
1
v
2
2
v
1
)e
3
(, V ) = div V =
1
v
1
+
2
v
2
+
3
v
3
(, V ) = 0
(V ) = V (, V )
(, aV ) = (a, V ) + a (, V )
aV = a V + aV
(, V U) = (V, U) (V, U)
Orthogonal transformations. Let U, V be vectors, i.e. they transform as the frame
vectors e
j
by means of the group O(X) . Then U V is a pseudovector (axial) vector,
i.e. A(U V ) = sgn (det A) (AU AV ) , A O(X) . A pseudovector is covariant for
the subgroup SO(X) and does not change under the symmetry x x. If U is a vector,
U is a pseudovector, then U V is a vector.
1
8.2 Maxwell equations
The electric eld E, the magnetic eld H, the electric induction D and the magnetic
induction B in the Euclidean space-time X R are related by the Maxwell system of
equations
H =
4
c
j +
1
c
D
t
(Amp`ere, Biot-Savart-Laplaces law) (8.1)
E =
1
c
B
t
(Faradays law) (8.2)
(, B) = 0 (Gausss law) (8.3)
(, D) = 4 (corollary of Coulombs law) (8.4)
with the sources: the charge density and the current j. The term D/t is called the
Maxwell displacement current. The Gauss units system - centimeter, gram, second - is
used; c 3 10
10
cm/ sec .
E, D are vector elds, i.e. they are covariant to the orthogonal group O(X) and
H, B are pseudovector eld (axial vectors), i.e. they are covariant to the special
orthogonal group SO(X) and do not change under the symmetry x x.
dimE = dimD = dimH = dimB = L
1/2
M
1/2
T
1
.
Integral form of the Maxwell system in the oriented space-time
_
S
(H, dl) =
4
c
_
S
(j, ds) +
1
c
t
_
S
(D, ds)
_
S
(E, dl) =
1
c
t
_
S
(B, ds)
_
U
(B, ds) = 0
_
U
(D, ds) = 4
_
U
dx
where
ds is the oriented surface element: ds = t
1
t
2
|ds| ; (t
1
, t
2
) is an orthonormal basis
of tangent elds in the surface S that dene the orientation of S;
dx is the volume form (not a density!) in X.
Conservation law for charge. The charge and the current are not arbitrary:
applying to the rst equation and
t
to the forth one, we get (, j) +
t
= 0 and in
the integral form
_
U
(j, ds) +
t
_
U
dx = 0
This is a conservation law for charge: if there is now current through the boundary U,
then the charge
_
U
dx is constant.
Symmetry. The system is invariant for the transformations:
E
= cos E + sin H, H
= cos H sin E
i.e. with respect to the group U (1) . This is a very simple example of gauge invariant
system. Another example: the Dirac-Maxwell system; the group is innite.
2
Potentials. The equation (2) with constant coecients can be solved:
B = A, E = A
0
1
c
A
t
A, A
0
are the vector and the scalar potentials. Physical sense: Aharonov-Bohm quantum
eect.
Material equations. To complete the Maxwell system one use material equations
D = D(E, H) , B = B(E, H) . In the simplest form:
D = E, B = H
is the (scalar) electric permittivity, is the (scalar) magnetic permeability. They are
dimensionless positive coecients depending on the medium; = = 1 for vacuum,
otherwise 1, 1. The velocity of electromagnetic waves is equal to v = c/
.
The principal symbol of the Maxwell system is the 8 6-matrix
1
=
_
_
_
_
I
3
I
3
0 (, )
(, ) 0
_
_
_
_
where = (
1
,
2
,
3
) and I
3
stands for the unit 33 matrix and = /c, = /c. There
are 28 6 6-minors. One of them is
det
_
_
_
_
_
_
_
_
0 0 0
0 0 0
0 0 0
0 0 0
0 0 0
0 0 0
_
_
_
_
_
_
_
_
=
2
(
2
2
)
2
,
where = (, , ) ,
2
=
2
+
2
+
2
, = ( )
1/2
.
Let A = C[, , , ] be the algebra of polynomials and J be the ideal generated by all
6 6-minors of
1
. We have J = (v
2
(x)
2
2
)
2
m
2
, where v
.
= ( )
1/2
is the velocity
of electro-magnetic waves in the medium and m A is the maximal ideal of the point
(0, 0) . Note that h = v
2
(x)
2
2
is the Hamiltonian function of the wave equation with
the velocity v. On the other hand, each component of the eld (E, H) satises the wave
equation with the principal symbol h(x; , ).
8.3 Harmonic analysis of solutions
Consider, rst, the wave equation in X R with a constant velocity v
_
2
t
2
v
2
_
u = 0
The symbol is
2
= h = v
2
2
. The characteristic variety is the cone {h(, ) = 0}
C
4
. A general solution is equal to a superposition of exponential solutions exp ( ((, x) + t));
the algebraic condition is that h(, ) = 0.
3
Theorem 1 Let be a convex open set in space-time. An arbitrary generalized solution
of the wave equation in can be written in the form
u(x) =
_
h=0
exp ( ((, x) + t)) m, (8.5)
where m is a complex-valued density supported by the variety {h = 0} such that for an
arbitrary compact K we have
_
exp (p
K
(Im(, )))
_
||
2
+||
2
+ 1
_
q
|m| <
for some q = q (K) . Vice versa, for any density that fulls this condition the integral (5)
is a generalized solution of the wave equation in .
The function p
K
is the Minkowski functional of K. The density m is not unique.
Maxwell system. Suppose that the coecients and are constant and j = 0, =
0. The plane waves
E = exp ( ((, x) + t)) e, H = exp ( ((, x) + t)) h (8.6)
If the vectors e, h satisfying
e + h = 0, e h = 0, (, h) = 0, (, e) = 0
then the plane wave (5) satises the Maxwell system in the free medium. Moreover, an
arbitrary solution is a superposition of the plane waves. Take the 6 6-matrix
_
_
(8.7)
=
_
_
_
_
_
_
_
_
2
0
2
2
0
2
2
0
0
2
2
0
2
0
2
2
_
_
_
_
_
_
_
_
Each line of this matrix satises (6), since V = ||
2
V + (, V ) .
Theorem 2 Let (e
j
, h
j
) , j = 1, 2 be arbitrary lines of the matrix (7) and be an
arbitrary convex domain in the space-time X R. An arbitrary generalized solution of
the Maxwell system without sources in can be written in the form
E =
_
h=0
exp ( ((, x) + t))
_
e
1
m
1
(, ) + e
2
m
2
(, )
,
H =
_
h=0
exp ( ((, x) + t))
_
h
1
m
1
(, ) + h
2
m
2
(, )
,
where m
1
, m
2
are some complex-valued densities supported in the variety {h = 0} such
that
_
exp (p
K
(Im(, )))
_
||
2
+||
2
+ 1
_
q
_
m
1
m
2
_
<
for an arbitrary compact set K and some constant q = q (K) .
4
8.4 Cauchy problem
Write the Maxwell system with sources:
j = 4c
1
j, = 4c
1
:
H
t
( E) =
j (8.8)
E +
t
( H) = 0
(, H) = 0
(, E) =
and variable coecients = (x) , = (x) . This is a overdetermined system: the
conservation law (, j) +
t
= 0 is a necessary condition for existence of a solution. The
system is hyperbolic in a sense; we can solve for the Cauchy problem for this system
E (x, 0) = E
0
(x) ,
t
E (x, 0) = E
1
(x) , H (x, 0) = H
0
(x) ,
t
H (x, 0) = H
1
(x)
provided more necessary conditions are satised:
H
0
E
1
= j (x, 0) , E
0
+ H
1
= 0,
( H
0
) = ( H
1
) = 0, (, E
0
) = (x, 0) , (, E
1
) =
t
(x, 0)
These equations together with the conservation law are the consistency conditions.
Theorem 3 Suppose that the coecients , are smooth functions in X and the sources
j, D
(X R) .
Proof. For unknown E, H we denote by F
i
, i = 1, 2, 3, 4 the left sides of the equations
(8) respectively. We nd
t
F
1
+
1
F
2
2
t
E +
_
1
E
_
=
t
t
F
2
+
1
F
1
2
t
H +
_
1
H
_
=
1
j
We have
1
E
1
(E +(, E)) +
1
(E)
=
1
_
E +
_
1
F
4
_
_
1
( , E)
__
+
1
(E)
Therefore
t
F
1
+
1
F
2
+
1
_
1
F
4
_
2
t
E
1
E
1
_
1
( , E)
_
+
1
E
=
t
j
1
( )
which implies the equation for the electric eld
2
t
E
1
E
1
_
1
(, E)
_
+
1
E = S
E
.
=
t
j
1
( )
5
The principal part is the wave operator with velocity since v = ( )
1/2
. The Cauchy
problem for this equation and initial data E
0
, E
1
has unique generalized solution E in
XR. Apply the operator (, ) to this equation and get by the consistency of the source
t
(, F
1
)
_
,
1
_
1
F
4
__
=
_
,
t
j
1
( )
_
= W ,
where
W
.
=
2
t
_
,
1
( )
_
On the other hand
t
(, F
1
)+
_
,
1
_
1
F
4
__
=
2
t
(, E)+
_
,
1
(, E)
_
= W (, E) = WF
4
hence W (F
4
) = 0. The function F
4
vanishes for t = 0 together with the rst time
derivative in virtue of the consistency conditions.
Lemma 4 The Cauchy problem for the operator W has no more than one solution
From the Lemma we conclude that F
4
= , which proves the forth equation.
Similarly we nd
t
F
2
+
1
F
1
1
_
1
F
3
_
2
t
H H
1
_
1
( , H)
_
+
1
H = S
H
.
=
_
,
1
j
_
This equation has the same principal part up to a scalar factor and we can solve the
Cauchy problem for initial data H
0
, H
1
. Arguing as above, we check that this solution
fulls the third equation. Then we have the system
t
F
1
+
1
F
2
= S
E
t
F
2
+
1
F
1
= S
H
Apply the operator
t
to the rst equation and the operator
1
to the second and
take the sum
2
t
F
1
+
1
_
1
F
1
_
=
t
S
E
+
1
S
H
(8.9)
=
t
_
j +
1
( )
_
+
1
__
,
1
j
__
We have
1
_
1
F
1
_
=
_
1
_
_
1
F
1
_
+
1
1
F
1
= ...
1
+
_
,
1
F
1
_
= ...
1
1
F
1
+
_
_
1
_
, F
1
_
+
_
1
(, F
1
)
_
and the last term vanishes since (, F
1
) = 0. Therefore the left side of (9) is equal to
UF
1
, where
U
.
=
2
t
1
1
+
_
1
_
_
1
_
+
_
_
1
_
,
_
6
The principal part is again the wave operator with the velocity v. The right side of (9)
is equal to U in virtue of the conservation law. Thus we have U (F
1
) = 0. We argue
as above and check the rst equation. The second one can veried in the same way.
Proof of Lemma. We will to show that Wu = 0 and u(x, 0) = u
t
(x, 0) = 0 implies
u = 0. Suppose for simplicity that u(, t) H
2
2
(X) for any value of time. Then we can
show the integral conservation law
t
_
_
u
2
t
+
1
|( u)|
2
_
dx = 2
_
u
t
_
u
tt
_
1
,
_
( u)
_
dx = 0
It follows that integral of
_
u
2
t
+
1
|( u)|
2
_
dx does not depend of time. It vanishes
for t = 0, hence vanishes for all times. To remove the assumption we continue u = 0 for
t < 0 and change the variables t
= t + |x x
0
|
2
, x
= x, where > 0, x
0
is arbitrary.
The function u has compact support in each hypersurface t
= for any .
8.5 Local conservation laws
The quadratic forms
E
2
= (E, E) , H
2
= (H, H) , S
.
=
v
4
E H
are called electric energy, magnetic energy and energy ux (Poynting vector), respec-
tively. We have dim(E
2
dx) = dim(H
2
dx) = dimSdx = M (L/T)
2
which equals the
dimension of energy.
Consider the Hamiltonian ow F generated by the function h. Its projection to XR
is the geodesic ow of the metric g = v
2
ds
2
.
Theorem 5 The densities E
2
dx, H
2
dx are equal and is preserved by the ow F in the
approximation of geometrical optics. The vector eld E is orthogonal to H and both are
orthogonal to any trajectory of F. Moreover the halfdensities
1/2
E
dx,
1/2
H
dx
keep parallel along any trajectory of F.
References
[1] P.Courant, D.Hilbert: Methods of Mathematical Physics
[2] V.Palamodov, Lecture Notes MP8
7